This is a first step for generating SSE rcp instructions for reciprocal
calcs when fast-math allows it. This is very similar to the rsqrt optimization
enabled in D5658 ( http://reviews.llvm.org/rL220570 ).
For now, be conservative and only enable this for AMD btver2 where performance
improves significantly both in terms of latency and throughput.
We may never enable this codegen for Intel Core* chips because the divider circuits
are just too fast. On SandyBridge, divss can be as fast as 10 cycles versus the 21
cycle critical path for the rcp + mul + sub + mul + add estimate.
Follow-on patches may allow configuration of the number of Newton-Raphson refinement
steps, add AVX512 support, and enable the optimization for more chips.
More background here: http://llvm.org/bugs/show_bug.cgi?id=21385
Differential Revision: http://reviews.llvm.org/D6175
llvm-svn: 221706
So DWARF5 specs out auto deduced return types as DW_TAG_unspecified_type
with DW_AT_name "auto", and GCC implements this somewhat, but it
presents a few problems to do this with Clang.
GCC's implementation only applies to member functions where the auto
return type isn't deduced immediately (ie: member functions of templates
or member functions defined out of line). In the common case of an
inline deduced return type function, GCC emits the DW_AT_type as the
deduced return type.
Currently GDB doesn't seem to behave too well with this debug info - it
treats the return type as 'void', even though the definition of the
function has the correctly deduced return type (I guess it sees the
return type the declaration has, doesn't understand it, and assumes
void). This means the function's ABI might be broken (non-trivial return
types, etc), etc.
Clang, on the other hand doesn't track this particular case of a
deducable return type that is deduced immediately versus one that is
deduced 'later'. So if we implement the DWARF5 representation, all
deducible return type functions would get adverse GDB behavior
(including deduced return type lambda functions, inline deduced return
type functions, etc).
Also, we can't just do this for auto types that are not deduced -
because Clang marks even the declaration's return type as deduced (&
provides the underlying type) once a definition is seen that allows the
deduction. So we have to ignore even deduced types - but we can't do
that for auto variables (because this representation only applies to
function declarations - variables and function definitions need the real
type so the function can be called, etc) so we'd need to add an extra
flag to the type unwrapping/creation code to indicate when we want to
see through deduced types and when we don't. It's also not as simple as
just checking at the top level when building a function type (for one
thing, we reuse the function type building for building function pointer
types which might also have 'auto' in them - but be the type of a
variable instead) because the auto might be arbitrarily deeply nested
("auto &", "auto (*)()", etc...)
So, with all that said, let's do the simple thing that works in existing
debuggers for now and treat these functions the same way we do function
templates and implicit special members: omit them from the member list,
since they can't be correctly called anyway (without knowing the return
type the ABI isn't know and a function call could put the arguments in
the wrong place) so they're not much use to the user.
At some point in the future, when GDB understands the DWARF5
representation better it might be worth plumbing through the extra type
builder handling to avoid looking through AutoType for some callers,
etc...
llvm-svn: 221704
My original support for the general dynamic and local dynamic TLS
models contained some fairly obtuse hacks to generate calls to
__tls_get_addr when lowering a TargetGlobalAddress. Rather than
generating real calls, special GET_TLS_ADDR nodes were used to wrap
the calls and only reveal them at assembly time. I attempted to
provide correct parameter and return values by chaining CopyToReg and
CopyFromReg nodes onto the GET_TLS_ADDR nodes, but this was also not
fully correct. Problems were seen with two back-to-back stores to TLS
variables, where the call sequences ended up overlapping with unhappy
results. Additionally, since these weren't real calls, the proper
register side effects of a call were not recorded, so clobbered values
were kept live across the calls.
The proper thing to do is to lower these into calls in the first
place. This is relatively straightforward; see the changes to
PPCTargetLowering::LowerGlobalTLSAddress() in PPCISelLowering.cpp.
The changes here are standard call lowering, except that we need to
track the fact that these calls will require a relocation. This is
done by adding a machine operand flag of MO_TLSLD or MO_TLSGD to the
TargetGlobalAddress operand that appears earlier in the sequence.
The calls to LowerCallTo() eventually find their way to
LowerCall_64SVR4() or LowerCall_32SVR4(), which call FinishCall(),
which calls PrepareCall(). In PrepareCall(), we detect the calls to
__tls_get_addr and immediately snag the TargetGlobalTLSAddress with
the annotated relocation information. This becomes an extra operand
on the call following the callee, which is expected for nodes of type
tlscall. We change the call opcode to CALL_TLS for this case. Back
in FinishCall(), we change it again to CALL_NOP_TLS for 64-bit only,
since we require a TOC-restore nop following the call for the 64-bit
ABIs.
During selection, patterns in PPCInstrInfo.td and PPCInstr64Bit.td
convert the CALL_TLS nodes into BL_TLS nodes, and convert the
CALL_NOP_TLS nodes into BL8_NOP_TLS nodes. This replaces the code
removed from PPCAsmPrinter.cpp, as the BL_TLS or BL8_NOP_TLS
nodes can now be emitted normally using their patterns and the
associated printTLSCall print method.
Finally, as a result of these changes, all references to get-tls-addr
in its various guises are no longer used, so they have been removed.
There are existing TLS tests to verify the changes haven't messed
anything up). I've added one new test that verifies that the problem
with the original code has been fixed.
llvm-svn: 221703
The ISel lowering for global TLS access in PIC mode was creating a pseudo
instruction that is later expanded to a call, but the code was not
setting the hasCalls flag in the MachineFrameInfo alongside the adjustsStack
flag. This caused some functions to be mistakenly recognized as leaf functions,
and this in turn affected the decision to eliminate the frame pointer.
With the fix, hasCalls is properly set and the leaf frame pointer is correctly
preserved.
llvm-svn: 221695
LLVM replaces the SelectionDAG pattern (xor (set_cc cc x y) 1) with
(set_cc !cc x y), which is only correct when the xor has type i1.
Instead, we should check that the constant operand to the xor is all
ones.
llvm-svn: 221693
an __unknown_anytype(...). In this case, we rebuild the
vararg function type specially to convert the call expression
to something that IRGen can handle. However, FunctionDecl
as rebuilt in RebuildUnknownAnyExpr::resolveDecl is bogus and
results in crash when accessing its params later on. This
patch fixes the crash by rebuilding the FunctionDecl to match
its new resolved type. rdar://15297105.
(patch reapplied after lldb issue was fixed in r221660).
llvm-svn: 221691
Summary:
The option -asan-instrument-assembly is declared in the X86 backend.
If I test on PowerPC configured with LLVM_TARGETS_TO_BUILD=All then the
option is tolerated but ignored.
If I test on PowerPC configured with LLVM_TARGETS_TO_BUILD=PowerPC then
the testsuite fails with:
[ 93%] Generating ASAN_INST_TEST_OBJECTS.gtest-all.cc.powerpc64-inline.o
clang (LLVM option parsing): Unknown command line argument '-asan-instrument-assembly'. Try: 'clang (LLVM option parsing) -help'
Fix this inconsistency by only adding the option if that toolchain was
built with the X86 backend included.
Reviewers: kcc, samsonov, eugenis
Reviewed By: eugenis
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6190
llvm-svn: 221687
Summary:
This patch enables code generation for the MIPS II target. Pre-Mips32
targets don't have the MUL instruction, so we add the correspondent
pattern that uses the MULT/MFLO combination in order to retrieve the
product.
This is WIP as we don't support code generation for select nodes due to
the lack of conditional-move instructions.
Reviewers: dsanders
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6150
llvm-svn: 221686
The canonical name when printing assembly is still $29. The reason is that
GAS does not accept "$hwr_ulr" at the moment.
This addresses the comments from r221307, which reverted the original
commit r221299.
llvm-svn: 221685
I went back and forth on removing this - and tried dropping it for
a few weeks. But when you're working at an assembly language, it
really is helpful to have this displayed to show where the current
pc is.
llvm-svn: 221682
The original commit r221299 was reverted in r221307. I removed the name
"hrw_ulr" ($29) from the original commit because two tests were failing.
llvm-svn: 221681
Referencing one symbol from another in the same section does not
generally require a relocation. However, the MS linker has a feature
called /INCREMENTAL which enables incremental links. It achieves this
by creating thunks to the actual function and redirecting all
relocations to point to the thunk.
This breaks down with the old scheme if you have a function which
references, say, itself. On x86_64, we would use %rip relative
addressing to reference the start of the function from out current
position. This would lead to miscompiles because other references might
reference the thunk instead, breaking function pointer equality.
This fixes PR21520.
llvm-svn: 221678
cost model for signed division by power of 2 was improved for AArch64.
The revision r218607 missed test case for Loop Vectorization.
Adding it in this revision.
Differential Revision: http://reviews.llvm.org/D6181
llvm-svn: 221674
This fixes an issue with matching trunc -> assertsext -> zext on x86-64, which would not zero the high 32-bits. See PR20494 for details.
Recommitting - This time, with a hopefully working test.
Differential Revision: http://reviews.llvm.org/D6128
llvm-svn: 221672
This adds const to a few methods that already return const references or
creates a const version when they reterun non-const references.
llvm-svn: 221666
For all threadprivate variables which have constructor/destructor emit call to void __kmpc_threadprivate_register(ident_t * <Current Location>, void *<Original Global Addr>, kmpc_ctor <Constructor>, kmpc_cctor NULL, kmpc_dtor <Destructor>);
In expressions all references to such variables are replaced by calls to void *__kmpc_threadprivate_cached(ident_t *<Current Location>, kmp_int32 <Current Thread Id>, void *<Original Global Addr>, size_t <Size of Data>, void ***<Pointer to autogenerated cache – array of private copies of threadprivate variable>);
Test test/OpenMP/threadprivate_codegen.cpp checks that codegen is correct. Also it checks that codegen is correct after serialization/deserialization and one of passes verifies debug info.
Differential Revision: http://reviews.llvm.org/D4002
llvm-svn: 221663
being asked about symbols it doesn't know about. If
it's asked about a symbol by mangled name and it finds
nothing, then it will try again with the demangled
base name.
llvm-svn: 221660
runtime. This eliminates potential confusion
when the compiler has to deal with these weird
types later on.
One day I'd like to actually generate the proper
templates, but this is not the day that I write
the parser code to do that.
<rdar://problem/18887634>
llvm-svn: 221658
AVX2 is available.
According to IACA, the new lowering has a throughput of 8 cycles instead of 13
with the previous one.
Althought this lowering kicks in some SPECs benchmarks, the performance
improvement was within the noise.
Correctness testing has been done for the whole range of uint32_t with the
following program:
uint4 v = (uint4) {0,1,2,3};
uint32_t i;
//Check correctness over entire range for uint4 -> float4 conversion
for( i = 0; i < 1U << (32-2); i++ )
{
float4 t = test(v);
float4 c = correct(v);
if( 0xf != _mm_movemask_ps( t == c ))
{
printf( "Error @ %vx: %vf vs. %vf\n", v, c, t);
return -1;
}
v += 4;
}
Where "correct" is the old lowering and "test" the new one.
The patch adds a test case for the two custom lowering instruction.
It also modifies the vector cost model, which is why cast.ll and uitofp.ll are
modified.
2009-02-26-MachineLICMBug.ll is also modified because we now hoist 7
instructions instead of 4 (3 more constant loads).
rdar://problem/18153096>
llvm-svn: 221657
Colin LeMahieu