Chapter 3 of the QPX manual states that, "Scalar floating-point load
instructions, defined in the Power ISA, cause a replication of the source data
across all elements of the target register." Thus, if we have a load followed
by a QPX splat (from the first lane), the splat is redundant. This adds a late
MI-level pass to remove the redundant splats in some of these cases
(specifically when both occur in the same basic block).
This optimization is scheduled just prior to post-RA scheduling. It can't happen
before anything that might replace the load with some already-computed quantity
(i.e. store-to-load forwarding).
llvm-svn: 265047
The test case added in r265023 is failing on ninja-x64-msvc-RA-centos6.
Update the test to make less specific assumptions on code generation.
llvm-svn: 265026
PPCSimplifyAddress contains this code:
IntegerType *OffsetTy = ((VT == MVT::i32) ? Type::getInt32Ty(*Context)
: Type::getInt64Ty(*Context));
to determine the type to be used for an index register, if one needs
to be created. However, the "VT" here is the type of the data being
loaded or stored, *not* the type of an address. This means that if
a data element of type i32 is accessed using an index that does not
not fit into 32 bits, a wrong address is computed here.
Note that PPCFastISel is only ever used on 64-bit currently, so the type
of an address is actually *always* MVT::i64. Other parts of the code,
even in this same PPCSimplifyAddress routine, already rely on that fact.
Thus, this patch changes the code to simply unconditionally use
Type::getInt64Ty(*Context) as OffsetTy.
llvm-svn: 265023
The fast isel pass currently emits a COPY_TO_REGCLASS node to convert
from a F4RC to a F8RC register class during conversion of a
floating-point number to integer. There is actually no support in the
common code instruction printers to emit COPY_TO_REGCLASS nodes, so the
PowerPC back-end has special code there to simply ignore
COPY_TO_REGCLASS.
This is correct *if and only if* the source and destination registers of
COPY_TO_REGCLASS are the same (except for the different register class).
But nothing guarantees this to be the case, and if the register
allocator does end up allocating source and destination to different
registers after all, the back-end simply generates incorrect code. I've
included a test case that shows such incorrect code generation.
However, it seems that COPY_TO_REGCLASS is actually not intended to be
used at the MI layer at all. It is used during SelectionDAG, but always
lowered to a plain COPY before emitting MI. Other back-end's fast isel
passes never emit COPY_TO_REGCLASS at all. I suspect it is simply wrong
for the PowerPC back-end to emit it here.
This patch changes the PowerPC back-end to directly emit COPY instead of
COPY_TO_REGCLASS and removes the special handling in the instruction
printers.
Differential Revision: http://reviews.llvm.org/D18605
llvm-svn: 265020
When dealing with complex<float>, and similar structures with two
single-precision floating-point numbers, especially when such things are being
passed around by value, we'll sometimes end up loading both float values by
extracting them from one 64-bit integer load. It looks like this:
t13: i64,ch = load<LD8[%ref.tmp]> t0, t6, undef:i64
t16: i64 = srl t13, Constant:i32<32>
t17: i32 = truncate t16
t18: f32 = bitcast t17
t19: i32 = truncate t13
t20: f32 = bitcast t19
The problem, especially before the P8 where those bitcasts aren't legal (and
get expanded via the stack), is that it would have been better to use two
floating-point loads directly. Here we add a target-specific DAGCombine to do
just that. In short, we turn:
ld 3, 0(5)
stw 3, -8(1)
rldicl 3, 3, 32, 32
stw 3, -4(1)
lfs 3, -4(1)
lfs 0, -8(1)
into:
lfs 3, 4(5)
lfs 0, 0(5)
llvm-svn: 264988
Instead of using two feature bits, one to indicate the availability of the
popcnt[dw] instructions, and another to indicate whether or not they're fast,
use a single enum. This allows more consistent control via target attribute
strings, and via Clang's command line.
llvm-svn: 264690
Minimum density for both optsize and non optsize are now options
-sparse-jump-table-density (default 10) for non optsize functions
-dense-jump-table-density (default 40) for optsize functions, which
matches the current default. This improves several benchmarks at google
at the cost of a small codesize increase. For code compiled with -Os,
the old behavior continues
llvm-svn: 264689
The A2 cores support the popcntw/popcntd instructions, but they're microcoded,
and slower than our default software emulation. Specifically, popcnt[dw] take
approximately 74 cycles, whereas our software emulation takes only 24-28
cycles.
I've added a new target feature to indicate a slow popcnt[dw], instead of just
removing the existing target feature from the a2/a2q processor models, because:
1. This allows us to return more accurate information via the TTI interface
(I recognize that this currently makes no practical difference)
2. Is hopefully easier to understand (it allows the core's features to match
its manual while still having the desired effect).
llvm-svn: 264600
Intrinsic::maxnum and Intrinsic::minnum, along with the associated libc
function calls (fmax[f], etc.) generally map to function calls after lowering.
For some vector types with QPX at least, however, we can legally lower these,
and we don't need to prohibit CTR-based loops on their account.
It turned out, however, that the logic that checked the opcodes associated with
intrinsics was broken (it would set the Opcode variable, but that variable was
later checked only if set for some otherwise-external function call.
This fixes the latter problem and adds the FMAX/MINNUM mappings.
llvm-svn: 264532
This patch corresponds to review:
http://reviews.llvm.org/D17711
It disables direct moves on these operations in 32-bit mode since the patterns
assume 64-bit registers. The final patch is slightly different from the
Phabricator review as the bitcast operations needed to be disabled in 32-bit
mode as well. This fixes PR26617.
llvm-svn: 264282
For fcmp, major concern about the following 6 cases is NaN result. The
comparison result consists of 4 bits, indicating lt, eq, gt and un (unordered),
only one of which will be set. The result is generated by fcmpu
instruction. However, bc instruction only inspects one of the first 3
bits, so when un is set, bc instruction may jump to to an undesired
place.
More specifically, if we expect an unordered comparison and un is set, we
expect to always go to true branch; in such case UEQ, UGT and ULT still
give false, which are undesired; but UNE, UGE, ULE happen to give true,
since they are tested by inspecting !eq, !lt, !gt, respectively.
Similarly, for ordered comparison, when un is set, we always expect the
result to be false. In such case OGT, OLT and OEQ is good, since they are
actually testing GT, LT, and EQ respectively, which are false. OGE, OLE
and ONE are tested through !lt, !gt and !eq, and these are true.
llvm-svn: 263753
This patch prevents CTR loops optimization when using soft float operations
inside loop body. Soft float operations use function calls, but function
calls are not allowed inside CTR optimized loops.
Patch by Aleksandar Beserminji.
Differential Revision: http://reviews.llvm.org/D17600
llvm-svn: 263727
This patch corresponds to review:
http://reviews.llvm.org/D17712
We were not clearing the TOC vector in PPCAsmPrinter when initializing it. This
caused duplicate definition asserts when the pass is reused on the module
(i.e. with -compile-twice or in JIT contexts).
llvm-svn: 263338
Corresponds to Phabricator review:
http://reviews.llvm.org/D16592
This fix includes both an update to how we handle the "generic" CPU on LE
systems as well as Anton's fix for the Fast Isel issue.
llvm-svn: 262233
Currently we always expand ISD::FNEG. For v4f32 and v2f64 vector types VSX has
native support for this opcode
Phabricator: http://reviews.llvm.org/D17647
llvm-svn: 262079
This is what was meant to be in the initial commit to fix this bug. The
parens were missing. This commit also adds a test case for the bug and
has undergone full testing on PPC and X86.
llvm-svn: 261546
Summary:
Without this, this command
$ llvm-run llc -stop-after machine-cp -o - <( echo '' )
outputs an error, because we close stdout twice -- once when closing the
file opened for "-o", and again when closing outs().
Also clarify in the outs() definition that you can't ever call it if you
want to open your own raw_fd_ostream on stdout.
Reviewers: jroelofs, tstellarAMD
Subscribers: jholewinski, qcolombet, dsanders, llvm-commits
Differential Revision: http://reviews.llvm.org/D17422
llvm-svn: 261286
Using the load immediate only when the immediate (whether signed or unsigned)
can fit in a 16-bit signed field. Namely, from -32768 to 32767 for signed and
0 to 65535 for unsigned. This patch also ensures that we sign-extend under the
right conditions.
llvm-svn: 259840
This patch consists of two parts: a performance fix in DAGCombiner.cpp
and a correctness fix in SelectionDAG.cpp.
The test case tests the bug that's uncovered by the performance fix, and
fixed by the correctness fix.
The performance fix keeps the containers required by the
hasPredecessorHelper (which is a lazy DFS) and reuse them. Since
hasPredecessorHelper is called in a loop, the overall efficiency reduced
from O(n^2) to O(n), where n is the number of SDNodes.
The correctness fix keeps iterating the neighbor list even if it's time
to early return. It will return after finishing adding all neighbors to
Worklist, so that no neighbors are discarded due to the original early
return.
llvm-svn: 259691
Recommited, after some fixing with test cases.
Updated test cases:
test/CodeGen/AArch64/arm64-misched-memdep-bug.ll
test/CodeGen/AArch64/tailcall_misched_graph.ll
Temporarily disabled test cases:
test/CodeGen/AMDGPU/split-vector-memoperand-offsets.ll
test/CodeGen/PowerPC/ppc64-fastcc.ll (partially updated)
test/CodeGen/PowerPC/vsx-fma-m.ll
test/CodeGen/PowerPC/vsx-fma-sp.ll
http://reviews.llvm.org/D8705
Reviewers: Hal Finkel, Andy Trick.
llvm-svn: 259673
The purpose of PPCVSXFMAMutate is to elide copies by changing FMA forms
on PPC.
%vreg6<def> = COPY %vreg96
%vreg6<def,tied1> = XSMADDASP %vreg6<tied0>, %vreg5<kill>, %vreg7
;v6 = v6 + v5 * v7
is replaced by
%vreg5<def,tied1> = XSMADDMSP %vreg5<tied0>, %vreg7, %vreg96
;v5 = v5 * v7 + v96
This was broken in the case where the target register was also used as a
multiplicand. Fix this case by checking for it and replacing both uses
with the copied register.
%vreg6<def> = COPY %vreg96
%vreg6<def,tied1> = XSMADDASP %vreg6<tied0>, %vreg5<kill>, %vreg6
;v6 = v6 + v5 * v6
is replaced by
%vreg5<def,tied1> = XSMADDMSP %vreg5<tied0>, %vreg96, %vreg96
;v5 = v5 * v96 + v96
llvm-svn: 259617
check that the sign extended constant fits into 16-bits if we want a
zero extended value, otherwise go ahead and put it together piecemeal.
Fixes PR26356.
llvm-svn: 259177
For historic reasons, the behavior of .align differs between targets.
Fortunately, there are alternatives, .p2align and .balign, which make the
interpretation of the parameter explicit, and which behave consistently across
targets.
This patch teaches MC to use .p2align instead of .align, so that people reading
code for multiple architectures don't have to remember which way each platform
does its .align directive.
Differential Revision: http://reviews.llvm.org/D16549
llvm-svn: 258750
The global entry point prologue currently assumes that the TOC
associated with a function is less than 2GB away from the function
entry point. This is always true when using the medium or small
code model, but may not be the case when using the large code model.
This patch adds a new variant of the ELFv2 global entry point prologue
that lifts the 2GB restriction when building with -mcmodel=large.
This works by emitting a quadword containing the distance from the
function entry point to its associated TOC immediately before the
entry point, and then using a prologue like:
ld r2,-8(r12)
add r2,r2,r12
Since creation of the entry point prologue is now split across two
separate routines (PPCLinuxAsmPrinter::EmitFunctionEntryLabel emits
the data word, PPCLinuxAsmPrinter::EmitFunctionBodyStart the prolog
code), I've switched to using named labels instead of just temporaries
to indicate the locations of the global and local entry points and the
new TOC offset data word.
These names are provided by new routines in PPCFunctionInfo modeled
after the existing PPCFunctionInfo::getPICOffsetSymbol.
Note that a corresponding change was committed to GCC here:
https://gcc.gnu.org/ml/gcc-patches/2015-12/msg00355.html
Reviewers: hfinkel
Differential Revision: http://reviews.llvm.org/D15500
llvm-svn: 257597
Only non-weighted predicates were handled in PPCInstrInfo::insertSelect. Handle
the weighted predicates as well.
This latent bug was triggered by r255398, because it added use of the
branch-weighted predicates.
While here, switch over an enum instead of an int to get the compiler to enforce
totality in the future.
llvm-svn: 257518
This is a fix for bug http://llvm.org/bugs/show_bug.cgi?id=25839.
For a PIC TLS variable access in a function, prologue (mflr followed by std and
stdu) gets scheduled after a tls_get_addr call. tls_get_addr messed up LR but
no one saves/restores it.
Also added a test for save/restore clobbered registers during calling __tls_get_addr.
Patch by Tim Shen
llvm-svn: 257137
This is the second in a set of patches for soft float support for ppc32,
it enables soft float operations.
Patch by Strahinja Petrovic.
Differential Revision: http://reviews.llvm.org/D13700
llvm-svn: 255516
This branch adds hints for highly biased branches on the PPC architecture. Even
in absence of profiling information, LLVM will mark code reaching unreachable
terminators and other exceptional control flow constructs as highly unlikely to
be reached.
Patch by Tom Jablin!
llvm-svn: 255398
Access to aligned globals gives us a chance to peephole optimize nonzero
offsets. If a struct is 4 byte aligned, then accesses to bytes 0-3 won't
overflow the available displacement. For example:
addis 3, 2, b4v@toc@ha
addi 4, 3, b4v@toc@l
lbz 5, b4v@toc@l(3) ; This is the result of the current peephole
lbz 6, 1(4) ; optimizer
lbz 7, 2(4)
lbz 8, 3(4)
If b4v is 4-byte aligned, we can skip using register 4 because we know
that b4v@toc@l+{1,2,3} won't overflow 32K, and instead generate:
addis 3, 2, b4v@toc@ha
lbz 4, b4v@toc@l(3)
lbz 5, b4v@toc@l+1(3)
lbz 6, b4v@toc@l+2(3)
lbz 7, b4v@toc@l+3(3)
Saving a register and an addition.
Larger alignments allow larger structures/arrays to be optimized.
llvm-svn: 255319
This was causing bad code gen and assembly that won't assemble, as
mixed altivec and vsx code would end up with a vsx high register
assigned to an altivec instruction, which won't work. Constraining the
classes allows the optimization to proceed.
llvm-svn: 255299
This patch corresponds to review:
http://reviews.llvm.org/D15286
LLVM IR frequently contains bitcast operations between floating point and
integer values of the same width. Doing this through memory operations is
quite expensive on PPC. This patch allows the use of direct register moves
between FPRs and GPRs for lowering bitcasts.
llvm-svn: 255246
AggressiveAntiDepBreaker was renaming registers specified by the user
for inline assembly. While this will work for compiler-specified
registers, it won't work for user-specified registers, and at the time
this runs, I don't currently see a way to distinguish them.
llvm-svn: 254532
