llvm-svn: 90009

the FIXME I added yesterday to be implemented.

llvm-svn: 90008

llvm-svn: 89982

the example in GCC PR16799 to:

LBB1_2:                                                     ## %bb1
	movl	%eax, %eax
	subq	%rax, %rdi
	movq	%rdi, (%rcx)
	movl	(%rdi), %eax
	testl	%eax, %eax
	je	LBB1_2

instead of:

LBB1_2:                                                     ## %bb1
	movl	(%rdi), %ecx
	subq	%rcx, %rdi
	movq	%rdi, (%rax)
	cmpl	$0, (%rdi)
	je	LBB1_2

llvm-svn: 89978

allows us to compile the example in readme.txt into:

LBB1_1:                                                     ## %bb
	movl	4(%rdx,%rax), %ecx
	movl	%ecx, %esi
	imull	(%rdx,%rax), %esi
	imull	%esi, %ecx
	movl	%esi, 8(%rdx,%rax)
	imull	%ecx, %esi
	movl	%ecx, 12(%rdx,%rax)
	movl	%esi, 16(%rdx,%rax)
	imull	%ecx, %esi
	movl	%esi, 20(%rdx,%rax)
	addq	$16, %rax
	cmpq	$4000, %rax
	jne	LBB1_1

instead of:

LBB1_1: 
	movl	(%rdx,%rax), %ecx
	imull	4(%rdx,%rax), %ecx
	movl	%ecx, 8(%rdx,%rax)
	imull	4(%rdx,%rax), %ecx
	movl	%ecx, 12(%rdx,%rax)
	imull	8(%rdx,%rax), %ecx
	movl	%ecx, 16(%rdx,%rax)
	imull	12(%rdx,%rax), %ecx
	movl	%ecx, 20(%rdx,%rax)
	addq	$16, %rax
	cmpq	$4000, %rax
	jne	LBB1_1

GCC (4.2) doesn't seem to be able to eliminate the loads in this 
testcase either, it generates:

L2:
	movl	(%rdx), %eax
	imull	4(%rdx), %eax
	movl	%eax, 8(%rdx)
	imull	4(%rdx), %eax
	movl	%eax, 12(%rdx)
	imull	8(%rdx), %eax
	movl	%eax, 16(%rdx)
	imull	12(%rdx), %eax
	movl	%eax, 20(%rdx)
	addl	$4, %ecx
	addq	$16, %rdx
	cmpl	$1002, %ecx
	jne	L2

llvm-svn: 89952

llvm-svn: 89951

llvm-svn: 89913

Rollback changes r89516: Added two SubtargetFeatures::AddFeatures methods, which accept a comma-separated string or already parsed command line parameters as input, and some code re-factoring to use these new methods.

llvm-svn: 89893

With the testcase for pr3120, the "threaded interpreter" runtime decreases
from 1788 to 1413 with this change.

llvm-svn: 89877

llvm-svn: 89873

llvm-svn: 89866

it is definitely profitable to tail duplicate indirect branches for x86.
This is likely to be true to various degrees for all modern x86 processors.

llvm-svn: 89865

llvm-svn: 89863

llvm-svn: 89850

using two swc/lwc instead of sdc/ldc.

llvm-svn: 89826

llvm-svn: 89824

unnecessary save/restore.

llvm-svn: 89823

llvm-svn: 89821

llvm-svn: 89819

Make tail duplication of indirect branches much more aggressive (for targets
that indicate that it is profitable), based on further experience with
this transformation.  I compiled 3 large applications with and without
this more aggressive tail duplication and measured minimal changes in code
size.  ("size" on Darwin seems to round the text size up to the nearest
page boundary, so I can only say that any code size increase was less than
one 4k page.) Radar 7421267.

llvm-svn: 89814

This violates the ABI (that area is "reserved"), and
while it is safe if all code is generated with current
compilers, there is some very old code around that uses
that slot for something else, and breaks if it is stored
into.  Adjust testcases looking for current behavior.
I've verified that the stack frame size is right in all
testcases, whether it changed or not.  7311323.

llvm-svn: 89811

llvm-svn: 89803

llvm-svn: 89748

No functional change.

llvm-svn: 89724

llvm-svn: 89723

than doing the same via constpool:
1. Load from constpool costs 3 cycles on A9, movt/movw pair - just 2.
2. Load from constpool might stall up to 300 cycles due to cache miss.
3. Movt/movw does not use load/store unit.
4. Less constpool entries => better compiler performance.

This is only enabled on ELF systems, since darwin does not have needed
relocations (yet).

llvm-svn: 89720

llvm-svn: 89718

way for each TargetJITInfo subclass to allocate its own stubs. This
means stubs aren't as exactly-sized anymore, but it lets us get rid of
TargetJITInfo::emitFunctionStubAtAddr(), which lets ARM and PPC
support the eager JIT, fixing http://llvm.org/PR4816.

* Rename the JITEmitter's stub creation functions to describe the kind
of stub they create. So far, all of them create lazy-compilation
stubs, but they sometimes get used when far-call stubs are needed.
Fixing http://llvm.org/PR5201 will involve fixing this.

llvm-svn: 89715

Note that "hasDotLocAndDotFile"-style debug info was already broken;
people wanting this functionality should implement it in the
AsmPrinter/DwarfWriter code.

llvm-svn: 89711

It's probably better in the long run to replace the
indirect-GlobalVariable system. That'll be done after a subsequent
patch.

llvm-svn: 89708

Massive refactoring of NEON instructions. Separate opcode from data size specifier suffix, move \t up stream to instruction format, and fix more 80 column violations.
This fixes the NEON asm printing so the "predicate" field is printed between the opcode and the data type suffix.

llvm-svn: 89706

llvm-svn: 89700

VDUPLND and VDUPLNQ to derive from N2V instead of N2VDup.  VDUPLND and VDUPLNQ
now expect op19_18 and op17_16 as the first two args.

llvm-svn: 89699

fold immediate of a + Const into the user as a subtract if it can fit as a negated two-part immediate.

llvm-svn: 89694

{?,?,?,?} as op11_8 for VEXTd and VEXTq.

llvm-svn: 89693

ARMInstrFormats.td and fixing VLD[234]LN* and VST[234]LN* to derive from NLdSt
instead of NLdStLN.

llvm-svn: 89684

should be left unspecified now that Bob Wilson has fixed pr5470.

llvm-svn: 89676

llvm-svn: 89672

llvm-svn: 89618

llvm-svn: 89576