x86 isel peeps.

llvm-svn: 92509

two labels as a truncate.

llvm-svn: 92455

llvm-svn: 92451

global variable initializer to require relocations.

llvm-svn: 92450

llvm-svn: 92407

 (X != null) | (Y != null) --> (X|Y) != 0
 (X == null) & (Y == null) --> (X|Y) == 0

so that instcombine can stop doing this for pointers.  This is part of PR3351,
which is a case where instcombine doing this for pointers (inserting ptrtoint)
is pessimizing code.

llvm-svn: 92406

llvm-svn: 92405

multiply sequence when the power is a constant integer.  Before, our
codegen for std::pow(.., int) always turned into a libcall, which was
really inefficient.

This should also make many gfortran programs happier I'd imagine.

llvm-svn: 92388

compare.  On other targets we end up with a call to memcmp because we don't
want 16 individual byte loads.  We should be able to use movups as well, but
we're failing to select the generated icmp.

llvm-svn: 92107

SDISel.  This optimization was causing simplifylibcalls to 
introduce type-unsafe nastiness.  This is the first step, I'll be 
expanding the memcmp optimizations shortly, covering things that
we really really wouldn't want simplifylibcalls to do.

llvm-svn: 92098

lowering code and update testcases.

llvm-svn: 91979

Remove target attribute break-sse-dep. Instead, do not fold load into sse partial update instructions unless optimizing for size.

llvm-svn: 91910

llvm-svn: 91717

be non-optimal. To be precise, we should avoid folding loads if the instructions
only update part of the destination register, and the non-updated part is not
needed. e.g. cvtss2sd, sqrtss. Unfolding the load from these instructions breaks
the partial register dependency and it can improve performance. e.g.

movss (%rdi), %xmm0
cvtss2sd %xmm0, %xmm0

instead of
cvtss2sd (%rdi), %xmm0

An alternative method to break dependency is to clear the register first. e.g.
xorps %xmm0, %xmm0
cvtss2sd (%rdi), %xmm0

llvm-svn: 91672

with unreachable.

llvm-svn: 91650

This protects this test from depending on codegen not performing the
tail call optimization by default.

llvm-svn: 91648

X86 instruction tables.

Also (while I was at it) cleaned up the X86 tables, removing tabs and
80-line violations.

This patch was reviewed by Chris Lattner, but please let me know if
there are any problems.

* X86*.td
	Removed tabs and fixed 80-line violations

* X86Instr64bit.td
	(IRET, POPCNT, BT_, LSL, SWPGS, PUSH_S, POP_S, L_S, SMSW)
		Added
	(CALL, CMOV) Added qualifiers
	(JMP) Added PC-relative jump instruction
	(POPFQ/PUSHFQ) Added qualifiers; renamed PUSHFQ to indicate
		that it is 64-bit only (ambiguous since it has no
		REX prefix)
	(MOV) Added rr form going the other way, which is encoded
		differently
	(MOV) Changed immediates to offsets, which is more correct;
		also fixed MOV64o64a to have to a 64-bit offset
	(MOV) Fixed qualifiers
	(MOV) Added debug-register and condition-register moves
	(MOVZX) Added more forms
	(ADC, SUB, SBB, AND, OR, XOR) Added reverse forms, which
		(as with MOV) are encoded differently
	(ROL) Made REX.W required
	(BT) Uncommented mr form for disassembly only
	(CVT__2__) Added several missing non-intrinsic forms
	(LXADD, XCHG) Reordered operands to make more sense for
		MRMSrcMem
	(XCHG) Added register-to-register forms
	(XADD, CMPXCHG, XCHG) Added non-locked forms
* X86InstrSSE.td
	(CVTSS2SI, COMISS, CVTTPS2DQ, CVTPS2PD, CVTPD2PS, MOVQ)
		Added
* X86InstrFPStack.td
	(COM_FST0, COMP_FST0, COM_FI, COM_FIP, FFREE, FNCLEX, FNOP,
	 FXAM, FLDL2T, FLDL2E, FLDPI, FLDLG2, FLDLN2, F2XM1, FYL2X,
	 FPTAN, FPATAN, FXTRACT, FPREM1, FDECSTP, FINCSTP, FPREM,
	 FYL2XP1, FSINCOS, FRNDINT, FSCALE, FCOMPP, FXSAVE,
	 FXRSTOR)
		Added
	(FCOM, FCOMP) Added qualifiers
	(FSTENV, FSAVE, FSTSW) Fixed opcode names
	(FNSTSW) Added implicit register operand
* X86InstrInfo.td
	(opaque512mem) Added for FXSAVE/FXRSTOR
	(offset8, offset16, offset32, offset64) Added for MOV
	(NOOPW, IRET, POPCNT, IN, BTC, BTR, BTS, LSL, INVLPG, STR,
	 LTR, PUSHFS, PUSHGS, POPFS, POPGS, LDS, LSS, LES, LFS,
	 LGS, VERR, VERW, SGDT, SIDT, SLDT, LGDT, LIDT, LLDT,
	 LODSD, OUTSB, OUTSW, OUTSD, HLT, RSM, FNINIT, CLC, STC,
	 CLI, STI, CLD, STD, CMC, CLTS, XLAT, WRMSR, RDMSR, RDPMC,
	 SMSW, LMSW, CPUID, INVD, WBINVD, INVEPT, INVVPID, VMCALL,
	 VMCLEAR, VMLAUNCH, VMRESUME, VMPTRLD, VMPTRST, VMREAD,
	 VMWRITE, VMXOFF, VMXON) Added
	(NOOPL, POPF, POPFD, PUSHF, PUSHFD) Added qualifier
	(JO, JNO, JB, JAE, JE, JNE, JBE, JA, JS, JNS, JP, JNP, JL,
	 JGE, JLE, JG, JCXZ) Added 32-bit forms
	(MOV) Changed some immediate forms to offset forms
	(MOV) Added reversed reg-reg forms, which are encoded
		differently
	(MOV) Added debug-register and condition-register moves
	(CMOV) Added qualifiers
	(AND, OR, XOR, ADC, SUB, SBB) Added reverse forms, like MOV
	(BT) Uncommented memory-register forms for disassembler
	(MOVSX, MOVZX) Added forms
	(XCHG, LXADD) Made operand order make sense for MRMSrcMem
	(XCHG) Added register-register forms
	(XADD, CMPXCHG) Added unlocked forms
* X86InstrMMX.td
	(MMX_MOVD, MMV_MOVQ) Added forms
* X86InstrInfo.cpp: Changed PUSHFQ to PUSHFQ64 to reflect table
	change

* X86RegisterInfo.td: Added debug and condition register sets
* x86-64-pic-3.ll: Fixed testcase to reflect call qualifier
* peep-test-3.ll: Fixed testcase to reflect test qualifier
* cmov.ll: Fixed testcase to reflect cmov qualifier
* loop-blocks.ll: Fixed testcase to reflect call qualifier
* x86-64-pic-11.ll: Fixed testcase to reflect call qualifier
* 2009-11-04-SubregCoalescingBug.ll: Fixed testcase to reflect call
  qualifier
* x86-64-pic-2.ll: Fixed testcase to reflect call qualifier
* live-out-reg-info.ll: Fixed testcase to reflect test qualifier
* tail-opts.ll: Fixed testcase to reflect call qualifiers
* x86-64-pic-10.ll: Fixed testcase to reflect call qualifier
* bss-pagealigned.ll: Fixed testcase to reflect call qualifier
* x86-64-pic-1.ll: Fixed testcase to reflect call qualifier
* widen_load-1.ll: Fixed testcase to reflect call qualifier

llvm-svn: 91638

llvm-svn: 91489

in local register allocator.  If a reg-reg copy has a phys reg
input and a virt reg output, and this is the last use of the phys
reg, assign the phys reg to the virt reg.  If a reg-reg copy has
a phys reg output and we need to reload its spilled input, reload
it directly into the phys reg than passing it through another reg.

Following 76208, there is sometimes no dependency between the def of
a phys reg and its use; this creates a window where that phys reg
can be used for spilling (this is true in linear scan also).  This
is bad and needs to be fixed a better way, although 76208 works too
well in practice to be reverted.  However, there should normally be
no spilling within inline asm blocks.  The patch here goes a long way
towards making this actually be true.

llvm-svn: 91485

llvm-svn: 91410

llvm-svn: 91405

1. Only perform (zext (shl (zext x), y)) -> (shl (zext x), y) when y is a constant. This makes sure it remove at least one zest.
2. If the shift is a left shift, make sure the original shift cannot shift out bits.

llvm-svn: 91399

llvm-svn: 91381

llvm-svn: 91378

llvm-svn: 91362

llvm-svn: 91223

llvm-svn: 91158

llvm-svn: 91148

aggregate return values. This fixes PR5754.

llvm-svn: 91145

This is used in some weird cases like general dynamic TLS model.
This fixes PR5723

llvm-svn: 91144

llvm-svn: 91105

It's not safe to coalesce a move where src and dst registers have different subregister indices. e.g.:
%reg16404:1<def> = MOV8rr %reg16412:2<kill>

llvm-svn: 91061

llvm-svn: 90988

vector_shuffle (scalar_to_vector (i32 load (ptr + 4))), undef, <0, 0, 0, 0>
=>
vector_shuffle (v4i32 load ptr), undef, <1, 1, 1, 1>

iff ptr is 16-byte aligned (or can be made into 16-byte aligned).

llvm-svn: 90984

Use FileCheck and set nounwind on calls.

llvm-svn: 90790

current form, it is too expensive in compile time.

llvm-svn: 90781

llvm-svn: 90653

The coalescer is supposed to clean these up, but when setting up parameters
for a function call, there may be copies to physregs. If the defining
instruction has been LICM'ed far away, the coalescer won't touch it.

The register allocation hint does not always work - when the register
allocator is backtracking, it clears the hints.

This patch takes care of a few more cases that r90163 missed.

llvm-svn: 90502

Don't pull vector sext through both hands of a logical operation, since doing so prevents the fusion of vector sext and setcc into vsetcc.
Add a testcase for the above transformation.
Fix a bogus use of APInt noticed while tracking this down.

llvm-svn: 90423

llvm-svn: 90352