Commit Graph

588 Commits

Author SHA1 Message Date
Chris Lattner 83263b8cfb Make X86TargetLowering::LowerSINT_TO_FP return without creating a dead
stack slot and store if the  SINT_TO_FP is actually legal.  This allows
us to compile:

double a(double b) {return (unsigned)b;}

to:

_a:
	cvttsd2siq	%xmm0, %rax
	movl	%eax, %eax
	cvtsi2sdq	%rax, %xmm0
	ret

instead of:

_a:
	subq	$8, %rsp
	cvttsd2siq	%xmm0, %rax
	movl	%eax, %eax
	cvtsi2sdq	%rax, %xmm0
	addq	$8, %rsp
	ret

crazy.

llvm-svn: 47660
2008-02-27 05:57:41 +00:00
Arnold Schwaighofer 3bfca3e942 Refactor according to Evan's and Anton's suggestions.
llvm-svn: 47635
2008-02-26 22:21:54 +00:00
Arnold Schwaighofer 1f17bf6171 Correct function comments.
llvm-svn: 47606
2008-02-26 17:50:59 +00:00
Arnold Schwaighofer 69a10f4112 Add support for intermodule tail calls on x86/32bit with
GOT-style position independent code. Before only tail calls to
protected/hidden functions within the same module were optimized.
Now all function calls are tail call optimized.

llvm-svn: 47594
2008-02-26 10:21:54 +00:00
Arnold Schwaighofer b01b99ec78 Change the lowering of arguments for tail call optimized
calls. Before arguments that could overwrite each other were
explicitly lowered to a stack slot, not giving the register allocator
a chance to optimize. Now a sequence of copyto/copyfrom virtual
registers ensures that arguments are loaded in (virtual) registers
before they are lowered to the stack slot (and might overwrite each
other). Also parameter stack slots are marked mutable for
(potentially) tail calling functions.

llvm-svn: 47593
2008-02-26 09:19:59 +00:00
Dale Johannesen 65b404d61c Revise previous patch per review.
llvm-svn: 47573
2008-02-25 22:29:22 +00:00
Dale Johannesen 32d84b1772 Expand removal of MMX memory copies to allow 1 level
of TokenFactor underneath chain (seems to be enough)

llvm-svn: 47554
2008-02-25 19:20:14 +00:00
Dale Johannesen 09f410b6d7 Split ParameterAttributes.h, putting the complicated
stuff into ParamAttrsList.h.  Per feedback from
ParamAttrs changes.

llvm-svn: 47504
2008-02-22 22:17:59 +00:00
Chris Lattner ab8bfc28c8 copy mmx values from/to memory with GPRs on x86-32
instead of with mmx registers.  This horribleness is apparently
done by gcc to avoid having to insert emms in places that really 
should have it.  This is the second half of rdar://5741668.

llvm-svn: 47474
2008-02-22 05:18:04 +00:00
Chris Lattner 997b3a65ca Start using GPR's to copy around mmx value instead of mmx regs.
GCC apparently does this, and code depends on not having to do
emms when this happens.  This is x86-64 only so far, second half
should handle x86-32.

rdar://5741668

llvm-svn: 47470
2008-02-22 02:09:43 +00:00
Anton Korobeynikov 40d67c59d5 Remove bunch of gcc 4.3-related warnings from Target
llvm-svn: 47369
2008-02-20 11:22:39 +00:00
Evan Cheng 6200c225e0 - When DAG combiner is folding a bit convert into a BUILD_VECTOR, it should check if it's essentially a SCALAR_TO_VECTOR. Avoid turning (v8i16) <10, u, u, u> to <10, 0, u, u, u, u, u, u>. Instead, simply convert it to a SCALAR_TO_VECTOR of the proper type.
- X86 now normalize SCALAR_TO_VECTOR to (BIT_CONVERT (v4i32 SCALAR_TO_VECTOR)). Get rid of X86ISD::S2VEC.

llvm-svn: 47290
2008-02-18 23:04:32 +00:00
Dan Gohman c589243107 Chris pointed out that it's not necessary to set i64 MUL to Expand
on x86-32 since i64 itself is not a Legal type. And, update some
comments.

llvm-svn: 47282
2008-02-18 19:34:53 +00:00
Dan Gohman a589ee11bb Don't mark scalar integer multiplication as Expand on x86, since x86
has plain one-result scalar integer multiplication instructions.
This avoids expanding such instructions into MUL_LOHI sequences that
must be special-cased at isel time, and avoids the problem with that
code that provented memory operands from being folded.

This fixes PR1874, addressesing the most common case. The uncommon
cases of optimizing multiply-high operations will require work
in DAGCombiner.

llvm-svn: 47277
2008-02-18 17:55:26 +00:00
Andrew Lenharth fedcf477b5 I cannot find a libgcc function for this builtin. Therefor expanding it to a noop (which is how it use to be treated). If someone who knows the x86 backend better than me could tell me how to get a lock prefix on an instruction, that would be nice to complete x86 support.
llvm-svn: 47213
2008-02-16 14:46:26 +00:00
Duncan Sands 4c95dbd69f In TargetLowering::LowerCallTo, don't assert that
the return value is zero-extended if it isn't
sign-extended.  It may also be any-extended.
Also, if a floating point value was returned
in a larger floating point type, pass 1 as the
second operand to FP_ROUND, which tells it
that all the precision is in the original type.
I think this is right but I could be wrong.
Finally, when doing libcalls, set isZExt on
a parameter if it is "unsigned".  Currently
isSExt is set when signed, and nothing is
set otherwise.  This should be right for all
calls to standard library routines.

llvm-svn: 47122
2008-02-14 17:28:50 +00:00
Nate Begeman 53e1b3f9d5 Change how FP immediates are handled.
1) ConstantFP is now expand by default
2) ConstantFP is not turned into TargetConstantFP during Legalize
   if it is legal.

This allows ConstantFP to be handled like Constant, allowing for 
targets that can encode FP immediates as MachineOperands.

As a bonus, fix up Itanium FP constants, which now correctly match,
and match more constants!  Hooray.

llvm-svn: 47121
2008-02-14 08:57:00 +00:00
Dan Gohman 9ca025f1dc Assigning an APInt to 0 with plain assignment gives it a one-bit
size. Initialize these APInts to properly-sized zero values.

llvm-svn: 47099
2008-02-13 23:07:24 +00:00
Dan Gohman e1d9ee66ed Simplify some logic in ComputeMaskedBits. And change ComputeMaskedBits
to pass the mask APInt by value, not by reference. 

llvm-svn: 47096
2008-02-13 22:28:48 +00:00
Dan Gohman f990faf23b Convert SelectionDAG::ComputeMaskedBits to use APInt instead of uint64_t.
Add an overload that supports the uint64_t interface for use by clients
that haven't been updated yet.

llvm-svn: 47039
2008-02-13 00:35:47 +00:00
Nate Begeman 8ef50214f0 SSE4.1 64b integer insert/extract pattern support
Move formats into the formats file

llvm-svn: 47035
2008-02-12 22:51:28 +00:00
Evan Cheng 4d8c98b8f9 Unbreak various insert_vector_elt and extract_vector_elt tests in presence of SSE4.
llvm-svn: 47001
2008-02-12 07:59:45 +00:00
Nate Begeman 2d77e8e446 Enable SSE4 codegen and pattern matching.
Add some notes to the README.

llvm-svn: 46949
2008-02-11 04:19:36 +00:00
Dan Gohman 3a4be0fdef Rename MRegisterInfo to TargetRegisterInfo.
llvm-svn: 46930
2008-02-10 18:45:23 +00:00
Dale Johannesen 36c2967d89 64-bit (MMX) vectors do not need restrictive alignment.
128-bit vectors need it only when SSE is on.

llvm-svn: 46890
2008-02-08 19:48:20 +00:00
Dan Gohman 7a55a94ba1 Avoid needlessly casting away const qualifiers.
llvm-svn: 46877
2008-02-08 03:29:40 +00:00
Dan Gohman 16d4bc3dc0 Follow Chris' suggestion; change the PseudoSourceValue accessors
to return pointers instead of references, since this is always what
is needed.

llvm-svn: 46857
2008-02-07 18:41:25 +00:00
Dan Gohman 63a8452e9c Add SourceValue information for outgoing argument stores on x86.
llvm-svn: 46854
2008-02-07 16:28:05 +00:00
Dan Gohman 2d489b5081 Re-apply the memory operand changes, with a fix for the static
initializer problem, a minor tweak to the way the
DAGISelEmitter finds load/store nodes, and a renaming of the
new PseudoSourceValue objects.

llvm-svn: 46827
2008-02-06 22:27:42 +00:00
Dale Johannesen d88f1d060e Implement sseregparm.
llvm-svn: 46764
2008-02-05 20:46:33 +00:00
Nick Lewycky f5b9938ef6 Don't use uninitialized values. Fixes vec_align.ll on X86 Linux.
llvm-svn: 46666
2008-02-02 08:29:58 +00:00
Evan Cheng efd142a920 SDIsel processes llvm.dbg.declare by recording the variable debug information descriptor and its corresponding stack frame index in MachineModuleInfo. This only works if the local variable is "homed" in the stack frame. It does not work for byval parameter, etc.
Added ISD::DECLARE node type to represent llvm.dbg.declare intrinsic. Now the intrinsic calls are lowered into a SDNode and lives on through out the codegen passes.
For now, since all the debugging information recording is done at isel time, when a ISD::DECLARE node is selected, it has the side effect of also recording the variable. This is a short term solution that should be fixed in time.

llvm-svn: 46659
2008-02-02 04:07:54 +00:00
Evan Cheng 27b32b87ed Revert 46556 and 46585. Dan please fix the PseudoSourceValue problem and re-commit.
llvm-svn: 46623
2008-01-31 21:00:00 +00:00
Dan Gohman ed346f2ed5 Avoid unnecessarily casting away const.
llvm-svn: 46590
2008-01-31 01:01:48 +00:00
Dan Gohman 9ba4d76816 Rename ISD::FLT_ROUNDS to ISD::FLT_ROUNDS_ to avoid conflicting
with the real FLT_ROUNDS (defined in <float.h>).

llvm-svn: 46587
2008-01-31 00:41:03 +00:00
Dan Gohman 3646fdda67 Create a new class, MemOperand, for describing memory references
in the backend. Introduce a new SDNode type, MemOperandSDNode, for
holding a MemOperand in the SelectionDAG IR, and add a MemOperand
list to MachineInstr, and code to manage them. Remove the offset
field from SrcValueSDNode; uses of SrcValueSDNode that were using
it are all all using MemOperandSDNode now.

Also, begin updating some getLoad and getStore calls to use the
PseudoSourceValue objects.

Most of this was written by Florian Brander, some
reorganization and updating to TOT by me.

llvm-svn: 46585
2008-01-31 00:25:39 +00:00
Evan Cheng 29cfb67e28 Even though InsertAtEndOfBasicBlock is an ugly hack it still deserves a proper name. Rename it to EmitInstrWithCustomInserter since it does not necessarily insert
instruction at the end.

llvm-svn: 46562
2008-01-30 18:18:23 +00:00
Evan Cheng 084a1cdcdd Work in progress. This patch *fixes* x86-64 calls which are modelled as StructRet but really should be return in registers, e.g. _Complex long double, some 128-bit aggregates. This is a short term solution that is necessary only because llvm, for now, cannot model i128 nor call's with multiple results.
Status: This only works for direct calls, and only the caller side is done. Disabled for now.

llvm-svn: 46527
2008-01-29 19:34:22 +00:00
Dale Johannesen 2b3bc30420 Handle 'X' constraint in asm's better.
llvm-svn: 46485
2008-01-29 02:21:21 +00:00
Chris Lattner d05d2011d0 Use fldz and fld1 for long double constants instead of a constant pool load.
llvm-svn: 46411
2008-01-27 06:19:31 +00:00
Chris Lattner 250789f1bd Remove some code for inferring alignment info from the x86 backend
now that the dag combiner does it.

llvm-svn: 46404
2008-01-26 20:07:42 +00:00
Chris Lattner f4523c35cb optimize fxor like for
llvm-svn: 46345
2008-01-25 06:14:17 +00:00
Chris Lattner 84ab724e06 Add target-specific dag combines for FAND(x,0) and FOR(x,0). This allows
us to compile:

double test(double X) {
  return copysign(0.0, X);
}

into:

_test:
	andpd	LCPI1_0(%rip), %xmm0
	ret

instead of:
_test:
	pxor	%xmm1, %xmm1
	andpd	LCPI1_0(%rip), %xmm1
	movapd	%xmm0, %xmm2
	andpd	LCPI1_1(%rip), %xmm2
	movapd	%xmm1, %xmm0
	orpd	%xmm2, %xmm0
	ret

llvm-svn: 46344
2008-01-25 05:46:26 +00:00
Chris Lattner a91f77eaac Significantly simplify and improve handling of FP function results on x86-32.
This case returns the value in ST(0) and then has to convert it to an SSE
register.  This causes significant codegen ugliness in some cases.  For 
example in the trivial fp-stack-direct-ret.ll testcase we used to generate:

_bar:
	subl	$28, %esp
	call	L_foo$stub
	fstpl	16(%esp)
	movsd	16(%esp), %xmm0
	movsd	%xmm0, 8(%esp)
	fldl	8(%esp)
	addl	$28, %esp
	ret

because we move the result of foo() into an XMM register, then have to
move it back for the return of bar.

Instead of hacking ever-more special cases into the call result lowering code
we take a much simpler approach: on x86-32, fp return is modeled as always 
returning into an f80 register which is then truncated to f32 or f64 as needed.
Similarly for a result, we model it as an extension to f80 + return.

This exposes the truncate and extensions to the dag combiner, allowing target
independent code to hack on them, eliminating them in this case.  This gives 
us this code for the example above:

_bar:
	subl	$12, %esp
	call	L_foo$stub
	addl	$12, %esp
	ret

The nasty aspect of this is that these conversions are not legal, but we want
the second pass of dag combiner (post-legalize) to be able to hack on them.
To handle this, we lie to legalize and say they are legal, then custom expand
them on entry to the isel pass (PreprocessForFPConvert).  This is gross, but
less gross than the code it is replacing :)

This also allows us to generate better code in several other cases.  For 
example on fp-stack-ret-conv.ll, we now generate:

_test:
	subl	$12, %esp
	call	L_foo$stub
	fstps	8(%esp)
	movl	16(%esp), %eax
	cvtss2sd	8(%esp), %xmm0
	movsd	%xmm0, (%eax)
	addl	$12, %esp
	ret

where before we produced (incidentally, the old bad code is identical to what
gcc produces):

_test:
	subl	$12, %esp
	call	L_foo$stub
	fstpl	(%esp)
	cvtsd2ss	(%esp), %xmm0
	cvtss2sd	%xmm0, %xmm0
	movl	16(%esp), %eax
	movsd	%xmm0, (%eax)
	addl	$12, %esp
	ret

Note that we generate slightly worse code on pr1505b.ll due to a scheduling 
deficiency that is unrelated to this patch.

llvm-svn: 46307
2008-01-24 08:07:48 +00:00
Evan Cheng 35abd840a6 Let each target decide byval alignment. For X86, it's 4-byte unless the aggregare contains SSE vector(s). For x86-64, it's max of 8 or alignment of the type.
llvm-svn: 46286
2008-01-23 23:17:41 +00:00
Duncan Sands 95d46ef887 The last pieces needed for loading arbitrary
precision integers.  This won't actually work
(and most of the code is dead) unless the new
legalization machinery is turned on.  While
there, I rationalized the handling of i1, and
removed some bogus (and unused) sextload patterns.
For i1, this could result in microscopically
better code for some architectures (not X86).
It might also result in worse code if annotating
with AssertZExt nodes turns out to be more harmful
than helpful.

llvm-svn: 46280
2008-01-23 20:39:46 +00:00
Chris Lattner 1ea55cf816 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 72733e573b * 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
Duncan Sands 32b0ff6814 Trampoline support for x86-64. This looks like
it should work, but I have no machine to test
it on.  Committed because it will at least
cause no harm, and maybe someone can test it
for me!

llvm-svn: 46098
2008-01-16 22:55:25 +00:00
Chris Lattner e8bb9f2190 make it more clear that this predicate only applies to scalar FP types.
llvm-svn: 46058
2008-01-16 06:24:21 +00:00