| Age | Commit message (Collapse) | Author |
|---|
|
|
|
first
|
|
Add portable support for concurrency in PNaCl's ABI:
- Promote volatile to atomic.
- Promote all memory ordering to sequential consistency.
- Rewrite all atomic operations to frozen NaCl intrinsics for pexe.
- Rewrite atomic intrinsics to LLVM instructions for translation.
This change also adds documentation to the PNaCl language reference, as
well as tests where it makes sense.
A future CL could clean up more of our code which mentions atomics,
volatiles, memory orderings.
Multiple reviewers because this is a big patch:
- eliben: LLVM-fu and ResolvePNaClIntrinsics.
- dschuff: ABI stability.
- mseaborn: ABI stability.
- sehr: Tron-duty (fight for the user's programs to work).
BUG= https://code.google.com/p/nativeclient/issues/detail?id=3475
R=dschuff@chromium.org, eliben@chromium.org, sehr@google.com
TEST= (cd ./pnacl/build/llvm_x86_64; ninja check-all) && ./pnacl/test.sh test-x86-32 && ./pnacl/test.sh test-x86-64 && ./pnacl/test.sh test-arm && ./pnacl/test.sh test-x86-32-sbtc && ./pnacl/test.sh test-x86-64-sbtc && ./pnacl/test.sh test-arm-sbtc
Review URL: https://codereview.chromium.org/17777004
|
|
The use of StripDeadPrototypes in PNaClABISimplify.cpp wasn't always
having an effect: it doesn't work if there are dead constant
references remaining to a function declaration. ReplacePtrsWithInts
was leaving some of these references behind, so fix it.
BUG=none
TEST=llvm-lit tests
Review URL: https://codereview.chromium.org/17636006
|
|
Disallow i1 on loads/stores and require the conversions to i8 to be
explicit. Add a pass, PromoteI1Ops, that adds the conversions.
(Load/store on i1 occur in practice in small_tests for some boolean
globals.)
Disallow i1 for most arithmetic/comparison operations since these
aren't very useful and it's a nuisance for a code generator to have to
support these. I haven't seen these occur in practice, but
PromoteI1Ops nevertheless expands them.
We still allow and/or/xor on i1 because these do occur in practice,
and they're less of a nuisance to handle because they never overflow:
no truncation to 1 bit is required, unlike with adds.
Restrict the type of alloca's argument. Clang always uses i32 here.
Disallow i1 in switch instructions. Clang doesn't generate i1
switches for booleans.
Move CopyLoadOrStoreAttrs() helper into a header to reuse.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3490
TEST=PNaCl toolchain trybots + GCC torture tests + Spec2k
Review URL: https://codereview.chromium.org/17356011
|
|
Simplify the set of "alloca" instructions the ABI verifier allows.
Before this change, we used i8 arrays, such as:
alloca [8 x i8]
After this change, we will just use i8 with an explicit size value, so
that becomes:
alloca i8, i32 8
Allocation of variable-length arrays will require an explicit multiply
instruction. This means that the code generator no longer has to
handle an implicit multiplication in "alloca", reducing the burden on
fast-and-simple code generators a little. This means the PNaCl ABI
doesn't need to specify whether alloca's implicit multiplication
checks for overflow.
This doesn't affect what the backend generates. See
lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp, which handles
constant-size AllocaInsts (by filling out StaticAllocaMap) and which
is uses for both -O2 (SelectionDAG) and -O0 (FastISel) translation.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343
TEST=*.ll tests + PNaCl toolchain trybots
Review URL: https://codereview.chromium.org/17631004
|
|
The normal form introduced by ReplacePtrsWithInts (as documented in
the comments) is intended to have the property that ptrtoint and
inttoptr only convert to/from i32, not other size types.
Using IRBuilder's CreateZExtOrTrunc() broke that, though, because it
performs some constant folding on global variable references. Fix
this by creating CastInsts directly.
I found this via the ABI checks I've been writing, which gave this
error when building the sandboxed translator:
non-i32 ptrtoint: %expanded1 = ptrtoint void ()* @ARMCompilationCallback to i8
LLVM ERROR: PNaCl ABI verification failed
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343
TEST=*.ll tests + tested full PNaCl build with ABI checks applied too
Review URL: https://codereview.chromium.org/15955012
|
|
These markers work in a similar way to llvm.lifetime.start/end, so we
should remove them for similar reasons: it's not very well defined how
one marker cancels out the effects of the other.
Arguably, invariant.start/end are less useful than lifetime.start/end.
They are ignored by the backend. They are generated in fewer places:
invariant.start is generated by Clang (at -O1 or higher) when a const
global is initialised with a non-POD initialiser. invariant.end is
apparently not generated at all.
Do the stripping in ReplacePtrsWithInts for consistency with the
existing lifetime.start/end stripping.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3443
TEST=PNaCl toolchain trybots
Review URL: https://codereview.chromium.org/15995004
|
|
This widens i1, i8 and i16 function arguments and return types.
Factor out RecreateFunction() helper function from existing PNaCl
passes since this is a reoccurring code fragment.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3342
TEST=*.ll tests + PNaCl toolchain trybots + GCC torture tests + LLVM test suite
Review URL: https://codereview.chromium.org/15971007
|
|
Running the LLVM test suite with the ReplacePtrsWithInts pass enabled
produced a single failure (in MultiSource/Applications/SPASS),
revealing a corner case in which a mixture of forward and backward
references plus a bitcast causes the pass to fail (see
@forwards_reference() in the test).
The problem was that we were doing replaceAllUsesWith() on a
placeholder value too early. RewriteMap was mapping a bitcast to a
placeholder P, but RewriteMap's reference to P didn't get updated by
P->replaceAllUsesWith() and P became a dangling pointer.
The fix is:
* Change convert() to strip off casts first, so that RewriteMap isn't
used for mapping casts to converted values.
* Defer the replaceAllUsesWith() calls until after creating all the
replacement instructions. This makes the pass more robust against
instruction ordering in the input.
This requires debug instrinsics to be updated in a separate pass,
because replaceAllUsesWith() doesn't work for references by
metadata nodes.
This also fixes some pathological corner cases of cyclic references in
unreachable blocks.
Fix indentation in one place.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343
TEST=replace-ptrs-with-ints.ll + LLVM test suite
Review URL: https://codereview.chromium.org/15761003
|
|
ReplacePtrsWithInts converts IR to a normal form in which functions
don't reference any aggregate pointer types and pointer types only
appear inside a few instructions.
Change BlockAddress::replaceUsesOfWithOnConstant() to handle changing
a function's type by replacing a function with a bitcast ConstantExpr
of a new function.
BUG=https://code.google.com/p/nativeclient/issues/detail?id=3343
TEST=replace-ptrs-with-ints.ll + PNaCl toolchain trybots, torture tests, etc.
Review URL: https://codereview.chromium.org/14262011
|
