to avoid undefined behavior on overflow, noticed by John Regehr.

llvm-svn: 109594

This is about 4x faster and smaller than the existing scalarization.

llvm-svn: 109566

~40% faster vector shl <4 x i32> on SSE 4.1  Larger improvements for smaller types coming in future patches.

For:

define <2 x i64> @shl(<4 x i32> %r, <4 x i32> %a) nounwind readnone ssp {
entry:
  %shl = shl <4 x i32> %r, %a                     ; <<4 x i32>> [#uses=1]
  %tmp2 = bitcast <4 x i32> %shl to <2 x i64>     ; <<2 x i64>> [#uses=1]
  ret <2 x i64> %tmp2
}

We get:

_shl:                                   ## @shl
	pslld	$23, %xmm1
	paddd	LCPI0_0, %xmm1
	cvttps2dq	%xmm1, %xmm1
	pmulld	%xmm1, %xmm0
	ret

Instead of:

_shl:                                   ## @shl
	pshufd	$3, %xmm0, %xmm2
	movd	%xmm2, %eax
	pshufd	$3, %xmm1, %xmm2
	movd	%xmm2, %ecx
	shll	%cl, %eax
	movd	%eax, %xmm2
	pshufd	$1, %xmm0, %xmm3
	movd	%xmm3, %eax
	pshufd	$1, %xmm1, %xmm3
	movd	%xmm3, %ecx
	shll	%cl, %eax
	movd	%eax, %xmm3
	punpckldq	%xmm2, %xmm3
	movd	%xmm0, %eax
	movd	%xmm1, %ecx
	shll	%cl, %eax
	movd	%eax, %xmm2
	movhlps	%xmm0, %xmm0
	movd	%xmm0, %eax
	movhlps	%xmm1, %xmm1
	movd	%xmm1, %ecx
	shll	%cl, %eax
	movd	%eax, %xmm0
	punpckldq	%xmm0, %xmm2
	movdqa	%xmm2, %xmm0
	punpckldq	%xmm3, %xmm0
	ret

llvm-svn: 109549

llvm-svn: 109494

subregister operands like this:

%reg1040:sub_32bit<def> = MOV32rm <fi#-2>, 1, %reg0, 0, %reg0, %reg1040<imp-def>; mem:LD4[FixedStack-2](align=8)

Make them return false when subreg operands are present. VirtRegRewriter is
making bad assumptions otherwise.

This fixes PR7713.

llvm-svn: 109489

with a too-big register class.

llvm-svn: 109488

llvm-svn: 109458

llvm-svn: 109456

llvm-svn: 109450

llvm-svn: 109448

we are using AVX and no AVX version of the desired intruction is present,
this is better for incremental dev (without fallbacks it's easier to spot
what's missing). Not sure this is the best hack thought (we can also disable
all HasSSE* predicates by dinamically marking them 'false' if AVX is present)

llvm-svn: 109434

This assumption is not satisfied due to global mergeing.
Workaround the issue by temporary disablinge mergeing of const globals.
Also, ignore LLVM "special" globals. This fixes PR7716

llvm-svn: 109423

llvm-svn: 109421

llvm-svn: 109373

llvm-svn: 109372

llvm-svn: 109359

appropriate for targets without detailed instruction iterineries.
The scheduler schedules for increased instruction level parallelism in
low register pressure situation; it schedules to reduce register pressure
when the register pressure becomes high.

On x86_64, this is a win for all tests in CFP2000. It also sped up 256.bzip2
by 16%.

llvm-svn: 109300

llvm-svn: 109295

function live in set. This will give us tGPR for Thumb1 and GPR otherwise,
so the copy will be spillable. rdar://8224931

llvm-svn: 109293

comments explaining why it was wrong.  8225024.

Fix the real problem in 8213383: the code that splits very large
blocks when no other place to put constants can be found was not
considering the case that the block contained a Thumb tablejump.

llvm-svn: 109282

  it's too late to start backing off aggressive latency scheduling when most
  of the registers are in use so the threshold should be a bit tighter.
- Correctly handle live out's and extract_subreg etc.
- Enable register pressure aware scheduling by default for hybrid scheduler.
  For ARM, this is almost always a win on # of instructions. It's runtime
  neutral for most of the tests. But for some kernels with high register
  pressure it can be a huge win. e.g. 464.h264ref reduced number of spills by
  54 and sped up by 20%.

llvm-svn: 109279

llvm-svn: 109276

llvm-svn: 109248

llvm-svn: 109224

llvm-svn: 109222

llvm-svn: 109207

llvm-svn: 109206

Add complete assembler support for FMA3 instructions, with descriptions and encodings taken from the AVX manual

llvm-svn: 109204

SSE, so we can't return floating point values if this
is disabled.  Detect this error for clang.

With SSE1 only, f64 is a problem; it can be done, but
neither llvm-gcc nor clang has ever generated correct
code for it.  Since nobody noticed this I think it's
OK to treat it as an error for now.

This also handles SSE-sized vectors of floating point.
8207686, 8204109.

llvm-svn: 109201

Fix some AVX instructions which didnt had HasAVX prefix. And also a problem with PINSRW, which was totally wrong because of a typo I introduced previously

llvm-svn: 109198

ARM/PPC/MSP430-specific code (which are the only targets that
implement the hook) can directly reference their target-specific
instrinfo classes.

llvm-svn: 109171

Add remaining AVX instructions (most of them dealing with GR64 destinations. This complete the assembler support for the general AVX ISA. But we still miss instructions from FMA3 and CLMUL specific feature flags, which are now the next step

llvm-svn: 109168

llvm-svn: 109167

This is probably not the best way to implement "Force LR to 
be spilled if the Thumb function size is > 2048." do this, 
it should use the branch shortening infrastructure, but I'm
just preserving functionality here.

llvm-svn: 109165

llvm-svn: 109154

rip out the implementation of X86InstrInfo::GetInstSizeInBytes.
The code being ripped out just implemented a copy and hacked up
version of the (old) instruction encoder, and is buggy and 
terrible in other ways.  Since "GetInstSizeInBytes" is really 
only there to support the JIT's "NeedsExactSize" hook (which
noone is using), just rip out the code.  I will rip out the
NeedsExactSize hook next.

This resolves rdar://7617809 - switch X86InstrInfo::GetInstSizeInBytes to use X86MCCodeEmitter

llvm-svn: 109149

llvm-svn: 109125

llvm-svn: 109092

llvm-svn: 109091

llvm-svn: 109090