[asan] more code to merge crash callbacks. Doesn't fully work yet, but allows to hold performance experiments

llvm-svn: 160361

When truncating a result of a vector that is split we need
to use the result of the split vector, and not re-split the dead node.

llvm-svn: 160357

llvm-svn: 160354

llvm-svn: 160352

llvm-svn: 160350

large immediates. Add dag combine logic to recover in case the large
immediates doesn't fit in cmp immediate operand field.

int foo(unsigned long l) {
  return (l>> 47) == 1;
}

we produce

  %shr.mask = and i64 %l, -140737488355328
  %cmp = icmp eq i64 %shr.mask, 140737488355328
  %conv = zext i1 %cmp to i32
  ret i32 %conv

which codegens to

movq    $0xffff800000000000,%rax
andq    %rdi,%rax
movq    $0x0000800000000000,%rcx
cmpq    %rcx,%rax
sete    %al
movzbl    %al,%eax
ret

TargetLowering::SimplifySetCC would transform
(X & -256) == 256 -> (X >> 8) == 1
if the immediate fails the isLegalICmpImmediate() test. For x86,
that's immediates which are not a signed 32-bit immediate.

Based on a patch by Eli Friedman.

PR10328
rdar://9758774

llvm-svn: 160346

Speculatively fix crashes by code inspection. Can't reproduce them yet.

llvm-svn: 160344

Some units tests crashed on a different platform.

llvm-svn: 160341

This places limits on CollectSubexprs to constrains the number of
reassociation possibilities. It limits the recursion depth and skips
over chains of nested recurrences outside the current loop.

Fixes PR13361. Although underlying SCEV behavior is still potentially bad.

llvm-svn: 160340

Define a 'null_frag' SDPatternOperator node, which if referenced in an
instruction Pattern, results in the pattern being collapsed to be as-if
'[]' had been specified instead. This allows supporting a multiclass
definition where some instaniations have ISel patterns associated and
others do not.

For example,
multiclass myMulti<RegisterClass rc, SDPatternOperator OpNode = null_frag> {
  def _x : myI<(outs rc:), (ins rc:), []>;
  def _r : myI<(outs rc:), (ins rc:), [(set rc:, (OpNode rc:))]>;
}

defm foo : myMulti<GRa, not>;
defm bar : myMulti<GRb>;

llvm-svn: 160333

llvm-svn: 160329

Defer checking for registers in the MC AsmMatcher until the after user-defined match classes have been checked.  This allows the creation of MatchClass's that are supersets of a register class.

llvm-svn: 160327

llvm-svn: 160325

llvm-svn: 160317

uint32_t hi(uint64_t res)
{
        uint_32t hi = res >> 32;
        return !hi;
}

llvm IR looks like this:
define i32 @hi(i64 %res) nounwind uwtable ssp {
entry:
  %lnot = icmp ult i64 %res, 4294967296
  %lnot.ext = zext i1 %lnot to i32
  ret i32 %lnot.ext
}

The optimizer has optimize away the right shift and truncate but the resulting
constant is too large to fit in the 32-bit immediate field. The resulting x86
code is worse as a result:
        movabsq $4294967296, %rax       ## imm = 0x100000000
        cmpq    %rax, %rdi
        sbbl    %eax, %eax
        andl    $1, %eax

This patch teaches the x86 lowering code to handle ult against a large immediate
with trailing zeros. It will issue a right shift and a truncate followed by
a comparison against a shifted immediate.
        shrq    $32, %rdi
        testl   %edi, %edi
        sete    %al
        movzbl  %al, %eax

It also handles a ugt comparison against a large immediate with trailing bits
set. i.e. X >  0x0ffffffff -> (X >> 32) >= 1

rdar://11866926

llvm-svn: 160312

llvm-svn: 160311

The first variant accepts immediate number as the second argument.
The second variant accepts register operand as the second argument.

llvm-svn: 160307

Make ComputeDemandedBits return a deterministic result when computing an AssertZext value.
In the added testcase the constant 55 was behind an AssertZext of type i1, and ComputeDemandedBits
reported that some of the bits were both known to be one and known to be zero.

Together with Michael Kuperstein <michael.m.kuperstein@intel.com>

llvm-svn: 160305

This reverts commit 4ea70107c5e51230e9e60f0bf58a0f74aa4885ea.

llvm-svn: 160303

This reverts commit 600f7a90f3eef4c5108179b43e27cfd9e5de7cdc.

llvm-svn: 160302

This reverts commit e3013202259ed1e006c21817c63cf25d75982721.

llvm-svn: 160301

This reverts commit 11d3457afcda7848448dd7f11b2ede6552ffb9ea.

llvm-svn: 160300

Revert "Target/AMDGPU/R600KernelParameters.cpp: Fix two includes, <llvm/IRBuilder.h> and <llvm/TypeBuilder.h>"

This reverts commit 0258a6bdd30802f5cc0e8e57c8e768fde2aef590.

llvm-svn: 160299

This reverts commit ebc934ba32ee71abbb8f0f2eb6a0fbaa613ba0d2.

llvm-svn: 160298

This reverts commit 29f28bc14ad5a907f5dc849f004fafeec0aab33a.

llvm-svn: 160297

Revert "Target/AMDGPU/AMDILIntrinsicInfo.cpp: Use llvm_unreachable() in nonreturn function, instead of assert(0)."

This reverts commit 4ba4acc1bc2561b944a571edbb6a2dc78e357dfe.

llvm-svn: 160296

This reverts commit fef4aa1b16fcf7a472559abbbcf4c1adc9eb5ca6.

llvm-svn: 160295

Make it always return APInts with the same bitwidth for the same ConstantRange bitwidth to simply clients

llvm-svn: 160294

llvm-svn: 160293

[asan] a bit more refactoring, addressed some of the style comments from chandlerc, partially implemented crash callback merging (under flag)

llvm-svn: 160290

MSVC's implementation of isalnum will assert on characters > 255, so we need to use an unsigned char to ensure the integer promotion happens properly.  This fixes an assert in debug builds with CodeGen\X86\utf8.ll

llvm-svn: 160286

[asan] refactor instrumentation to allow merging the crash callbacks (not fully implemented yet, no functionality change except the BB order)

llvm-svn: 160284

llvm-svn: 160280

Target/AMDGPU/AMDILIntrinsicInfo.cpp: Use llvm_unreachable() in nonreturn function, instead of assert(0).

llvm-svn: 160279

llvm-svn: 160278

Mips shift instructions DSLL, DSRL and DSRA are transformed into
DSLL32, DSRL32 and DSRA32 respectively if the shift amount is between
32 and 63

Here is a description of DSLL:

Purpose: Doubleword Shift Left Logical Plus 32
To execute a left-shift of a doubleword by a fixed amount--32 to 63 bits

Description: GPR[rd] <- GPR[rt] << (sa+32)

The 64-bit doubleword contents of GPR rt are shifted left, inserting
 zeros into the emptied bits; the result is placed in
GPR rd. The bit-shift amount in the range 0 to 31 is specified by sa.

This patch implements the direct object output of these instructions.

llvm-svn: 160277

llvm-svn: 160276

Target/AMDGPU/R600KernelParameters.cpp: Fix two includes, <llvm/IRBuilder.h> and <llvm/TypeBuilder.h>

llvm-svn: 160275

1. FileCheck-ize epilogue.ll and allow another asm instruction to restore %rsp.
2. Remove check in widen_arith-3.ll that was hitting instruction in epilogue instead of
vector add.

llvm-svn: 160274

llvm-svn: 160273