For subtargets that use the custom lowering for divmod, e.g. gnueabi,
we used to check if the subtarget has hardware divide and then lower to
a div-mul-sub sequence if true, or to a libcall if false.
However, judging by the usage of hasDivide vs hasDivideInARMMode, it
seems that hasDivide only refers to Thumb. For instance, in the
ARMTargetLowering constructor, the code that specifies whether to use
libcalls for (S|U)DIV looks like this:
bool hasDivide = Subtarget->isThumb() ? Subtarget->hasDivide()
: Subtarget->hasDivideInARMMode();
In the case of divmod for arm-gnueabi, using only hasDivide() to
determine what to do means that instead of lowering to __aeabi_idivmod
to get the remainder, we lower to div-mul-sub and then further lower the
div to __aeabi_idiv. Even worse, if we have hardware divide in ARM but
not in Thumb, we generate a libcall instead of using it (this is not an
issue in practice since AFAICT none of the cores that we support have
hardware divide in ARM but not Thumb).
This patch fixes the code dealing with custom lowering to take into
account the mode (Thumb or ARM) when deciding whether or not hardware
division is available.
Differential Revision: https://reviews.llvm.org/D32005
llvm-svn: 300536
Use the same handling in the generic legalizer code as for the other
libcalls (G_FREM, G_FPOW).
Enable it on ARM for float and double so we can test it.
llvm-svn: 299931
A fix for the bug reported in PR30911.
The issue arises when multiple CALLSEQ_BEGIN nodes are unscheduled as
the last node to be unscheduled will gain access to the CallResource
register. But when a node is being picked, only CALLSEQ_END nodes are
checked against the CallResource and have their chains evaluated.
This then means that other CALLSEQ_BEGIN nodes can be scheduled
before the existing call sequence has been finalised. This patch adds
a check against the FrameSetup nodes in DelayForLiveRegs to prevent
this from happening.
Differential Revision: https://reviews.llvm.org/D31536
llvm-svn: 299926
BIC is generally faster, and it can put the output in a different
register from the input.
We already do this in Thumb2 mode; not sure why the equivalent fix
never got applied to ARM mode.
Differential Revision: https://reviews.llvm.org/D31797
llvm-svn: 299803
It turns out -float-abi=hard doesn't set the hard float calling
convention for libcalls. We need to use a hard float triple instead
(e.g. gnueabihf).
llvm-svn: 299761
Legalize to a libcall.
On this occasion, also start allowing soft float subtargets. For the
moment G_FREM is the only legal floating point operation for them.
llvm-svn: 299753
The new codepath has been in the tree for years, and there isn't any
reason to use two codepaths here.
Differential Revision: https://reviews.llvm.org/D30596
llvm-svn: 299723
In LowerMUL, the chain information is not preserved for the new
created Load SDNode.
For example, if a Store alias with one of the operand of Mul.
The Load for that operand need to be scheduled before the Store.
The dependence is recorded in the chain of Store, in TokenFactor.
However, when lowering MUL, the SDNodes for the new Loads for
VMULL are not updated in the TokenFactor for the Store. Thus the
chain is not preserved for the lowered VMULL.
llvm-svn: 299701
Summary:
Host CPU detection now supports Kryo, so we need to recognize it in ARM
target.
Reviewers: mcrosier, t.p.northover, rengolin, echristo, srhines
Reviewed By: t.p.northover, echristo
Subscribers: aemerson
Differential Revision: https://reviews.llvm.org/D31775
llvm-svn: 299674
This is a generic combine enabled via target hook to reduce icmp logic as discussed in:
https://bugs.llvm.org/show_bug.cgi?id=32401
It's likely that other targets will want to enable this hook for scalar transforms,
and there are probably other patterns that can use bitwise logic to reduce comparisons.
Note that we are missing an IR canonicalization for these patterns, and we will probably
prefer the pair-of-compares form in IR (shorter, more likely to fold).
Differential Revision: https://reviews.llvm.org/D31483
llvm-svn: 299542
This patch enables schedulers to specify instructions that
cannot be issued with any other instructions.
It also fixes BeginGroup/EndGroup.
Reviewed by: Andrew Trick
Differential Revision: https://reviews.llvm.org/D30744
llvm-svn: 298885
Summary:
The true and false operands for the CMOV are operands 0 and 1.
ARMISelLowering.cpp::computeKnownBits was looking at operands 1 and 2
instead. This can cause CMOV instructions to be incorrectly folded into
BFI if value set by the CMOV is another CMOV, whose known bits are
computed incorrectly.
This patch fixes the issue and adds a test case.
Reviewers: kristof.beyls, jmolloy
Subscribers: llvm-commits, aemerson, srhines, rengolin
Differential Revision: https://reviews.llvm.org/D31265
llvm-svn: 298624
The special case of zero sized values was previously not handled correctly.
This patch handles this by not promoting if the size is zero.
Patch by Tim Neumann.
Differential Revision: https://reviews.llvm.org/D31116
llvm-svn: 298320
This commit adds a parameter that lets us pass in the calling convention
of the call to CallLowering::lowerCall. This allows us to handle
situations where the calling convetion of the callee is different from
that of the caller.
Differential Revision: https://reviews.llvm.org/D31039
llvm-svn: 298254
Folding instructions when selecting can cause them to become dead.
Don't select these dead instructions (if they don't have other side
effects, and don't define physical registers).
Preserve existing tests by adding COPYs.
In some tests, the G_CONSTANT vregs never get constrained to a class:
the only use of the vreg was folded into another instruction, so the
G_CONSTANT, now dead, never gets selected.
llvm-svn: 298224
Handle TokenFactors more aggressively in
SDValue::reachesChainWithoutSideEffects. This isn't really a
very effective change anymore because of other changes to
chain handling, but it's a cheap check, and the expanded
comments are still useful.
It might be possible to loosen the hasOneUse() requirement with a
deeper analysis, but a naive implementation of that check would be
expensive.
Differential Revision: https://reviews.llvm.org/D29845
llvm-svn: 298156
This allows the optimization to rearrange loads and stores more
aggressively. This doesn't really affect performance, but it helps
codesize.
Differential Revision: https://reviews.llvm.org/D30839
llvm-svn: 298021
Citing http://bugs.llvm.org/show_bug.cgi?id=32288
The DWARF generated by LLVM includes this location:
0x55 0x93 0x04 DW_OP_reg5 DW_OP_piece(4) When GCC's DWARF is simply
0x55 (DW_OP_reg5) without the DW_OP_piece. I believe it's reasonable
to assume the DWARF consumer knows which part of a register
logically holds the value (low bytes, high bytes, how many bytes,
etc) for a primitive value like an integer.
This patch gets rid of the redundant DW_OP_piece when a subregister is
at offset 0. It also adds previously missing subregister masking when
a subregister is followed by another operation.
(This reapplies r297960 with two additional testcase updates).
rdar://problem/31069390
https://reviews.llvm.org/D31010
llvm-svn: 297965
Don't scalarize VSELECT->SETCC when operands/results needs to be widened,
or when the type of the SETCC operands are different from those of the VSELECT.
(VSELECT SETCC) and (VSELECT (AND/OR/XOR (SETCC,SETCC))) are handled.
The previous splitting of VSELECT->SETCC in DAGCombiner::visitVSELECT() is
no longer needed and has been removed.
Updated tests:
test/CodeGen/ARM/vuzp.ll
test/CodeGen/NVPTX/f16x2-instructions.ll
test/CodeGen/X86/2011-10-19-widen_vselect.ll
test/CodeGen/X86/2011-10-21-widen-cmp.ll
test/CodeGen/X86/psubus.ll
test/CodeGen/X86/vselect-pcmp.ll
Review: Eli Friedman, Simon Pilgrim
https://reviews.llvm.org/D29489
llvm-svn: 297930
If we got unlucky with register allocation and actual constpool placement, we
could end up producing a tTBB_JT with an index that's already been clobbered.
Technically, we might be able to fix this situation up with a MOV, but I think
the constant islands pass is complex enough without having to deal with more
weird edge-cases.
llvm-svn: 297871
Enable the selection of the 64-bit signed multiply accumulate
instructions which operate on 16-bit operands. These are enabled for
ARMv5TE onwards for ARM and for V6T2 and other DSP enabled Thumb
architectures.
Differential Revision: https://reviews.llvm.org/D30044
llvm-svn: 297809
Create nodes for smulwb and smulwt and move their selection from
DAGToDAG to DAG combine. smlawb and smlawt can then be selected
using tablegen. Added some helper functions to detect shift patterns
as well as a wrapper around SimplifyDemandBits. Added a couple of
extra tests.
Differential Revision: https://reviews.llvm.org/D30708
llvm-svn: 297716
Recommiting with compiler time improvements
Recommitting after fixup of 32-bit aliasing sign offset bug in DAGCombiner.
* Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search and chain alias analysis which only
checks for parallel stores through the chain subgraph. This is cleaner
as the separation of non-interfering loads/stores from the
store-merging logic.
When merging stores search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited.
This improves the quality of the output SelectionDAG and the output
Codegen (save perhaps for some ARM cases where we correctly constructs
wider loads, but then promotes them to float operations which appear
but requires more expensive constant generation).
Some minor peephole optimizations to deal with improved SubDAG shapes (listed below)
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the chain aggregation in the merged stores across code
paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seems sufficient to not cause regressions in
tests.
5. Remove Chain dependencies of Memory operations on CopyfromReg
nodes as these are captured by data dependence
6. Forward loads-store values through tokenfactors containing
{CopyToReg,CopyFromReg} Values.
7. Peephole to convert buildvector of extract_vector_elt to
extract_subvector if possible (see
CodeGen/AArch64/store-merge.ll)
8. Store merging for the ARM target is restricted to 32-bit as
some in some contexts invalid 64-bit operations are being
generated. This can be removed once appropriate checks are
added.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable, improving load-store forwarding. One test in
particular is worth noting:
CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store
forwarding converts a load-store pair into a parallel store and
a memory-realized bitcast of the same value. However, because we
lose the sharing of the explicit and implicit store values we
must create another local store. A similar transformation
happens before SelectionDAG as well.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
llvm-svn: 297695
We used to hit an unreachable in getRegBankFromRegClass when dealing with the
stack pointer. This commit adds support for the GPRsp reg class.
llvm-svn: 297621
This helps in cases involving bitfields where an AND is exposed by
legalization.
Differential Revision: https://reviews.llvm.org/D30472
llvm-svn: 297249
We cannot leave the identity copies 'select true, arg, undef' that this pass
inserts for arguments to simplify handling of values on swifterror arguments.
swifterror arguments have restrictions on their uses.
rdar://30839288
llvm-svn: 297197
The original patch r296865 was reverted as it broke the chromium builds for
Android https://bugs.llvm.org/show_bug.cgi?id=32134, this patch reapplies
r296865 with a fix to make sure it doesn't cause the build regression.
The problem was that intrinsic selection on int_arm_get_fpscr was failing in
ISel this was because the code to manually select this intrinsic still thought
it was the version with no side-effects (INTRINSIC_WO_CHAIN) which is wrong as
it doesn't semantically match the definition in the tablegen code which says it
does have side-effects, I've fixed this by updating the intrinsic type to
INTRINSIC_W_CHAIN (has side-effects). I've also added a test for this based on
Hans original reproducer.
Differential Revision: https://reviews.llvm.org/D30645
llvm-svn: 297137
Summary: Previously, it had always been materialized as a push/pop sequence.
Reviewers: labrinea, jroelofs
Reviewed By: jroelofs
Subscribers: llvm-commits, rengolin
Differential Revision: https://reviews.llvm.org/D30648
llvm-svn: 297134
A bit more painful than G_INSERT because it was more widely used, but this
should simplify the handling of extract operations in most locations.
llvm-svn: 297100
The intrinsics __builtin_arm_get_fpscr and __builtin_arm_set_fpscr read and
write to the fpscr (Floating-Point Status and Control Register) register.
A bug exists in the __builtin_arm_get_fpscr intrinsic definition in llvm which
treats this intrinsic as a IntroNoMem which means it's not a memory access and
doesn't have any other side-effects. Having this property on this intrinsic
means that various optimizations can be done on this such as common
sub-expression elimination with other reads. This can cause issues if there has
been write to this register, e.g.
void foo(int *p) {
p[0] = __builtin_arm_get_fpscr();
__builtin_arm_set_fpscr(1);
p[1] = __builtin_arm_get_fpscr();
}
in the above example the second read is currently CSE'd into the first read,
this is because llvm isn't aware that the write done by __builtin_arm_set_fpscr
effects the same register that __builtin_arm_get_fpscr reads from, to fix this
problem I've removed the property IntrNoMem so that __builtin_arm_get_fpscr is
treated as a memory access.
Differential Revision: https://reviews.llvm.org/D30542
llvm-svn: 296865
This patch causes compile times for some patterns to explode. I have
a (large, unreduced) test case that slows down by more than 20x and
several test cases slow down by 2x. I'm sending some of the test cases
directly to Nirav and following up with more details in the review log,
but this should unblock anyone else hitting this.
llvm-svn: 296862
In ARMPreAllocLoadStoreOpt::RescheduleOps, LastOp should be the last
operation which we want to merge. If we break out of the loop because
an operation has the wrong offset, we shouldn't use that operation
as LastOp.
This patch fixes some cases where we would move stores to the wrong
insert point.
Re-commit with a fix to increment NumMove in the right place.
Differential Revision: https://reviews.llvm.org/D30124
llvm-svn: 296815
This bug was introduced with:
https://reviews.llvm.org/rL296699
There may be a way to loosen the restriction, but for now just bail out
on any opaque constant.
The tests show that opacity is target-specific. This goes back to cost
calculations in ConstantHoisting based on TTI->getIntImmCost().
llvm-svn: 296768
Original commit message:
[ARM] Fix insert point for store rescheduling.
In ARMPreAllocLoadStoreOpt::RescheduleOps, LastOp should be the last
operation which we want to merge. If we break out of the loop because
an operation has the wrong offset, we shouldn't use that operation as
LastOp.
This patch fixes some cases where we would sink stores for no reason.
llvm-svn: 296718
In ARMPreAllocLoadStoreOpt::RescheduleOps, LastOp should be the last
operation which we want to merge. If we break out of the loop because
an operation has the wrong offset, we shouldn't use that operation as
LastOp.
This patch fixes some cases where we would sink stores for no reason.
Differential Revision: https://reviews.llvm.org/D30124
llvm-svn: 296708
This code starts from the high end of the sorted vector of offsets, and
works backwards: it tries to find contiguous offsets, process them, then
pops them from the end of the vector. Most of the code agrees with this
order of processing, but one loop doesn't: it instead processes elements
from the low end of the vector (which are nodes with unrelated offsets).
Fix that loop to process the correct elements.
This has a few implications. One, we don't incorrectly return early when
processing multiple groups of offsets in the same block (which allows
rescheduling prera-ldst-insertpt.mir). Two, we pick the correct insert
point for loads, so they're correctly sorted (which affects the
scheduling of vldm-liveness.ll). I think it might also impact some of
the heuristics slightly.
Differential Revision: https://reviews.llvm.org/D30368
llvm-svn: 296701
This is part of the ongoing attempt to improve select codegen for all targets and select
canonicalization in IR (see D24480 for more background). The transform is a subset of what
is done in InstCombine's FoldOpIntoSelect().
I first noticed a regression in the x86 avx512-insert-extract.ll tests with a patch that
hopes to convert more selects to basic math ops. This appears to be a general missing DAG
transform though, so I added tests for all standard binops in rL296621
(PowerPC was chosen semi-randomly; it has scripted FileCheck support, but so do ARM and x86).
The poor output for "sel_constants_shl_constant" is tracked with:
https://bugs.llvm.org/show_bug.cgi?id=32105
Differential Revision: https://reviews.llvm.org/D30502
llvm-svn: 296699
Summary:
Avoids tons of prologue boilerplate when arguments are passed in memory
and left in memory. This can happen in a debug build or in a release
build when an argument alloca is escaped. This will dramatically affect
the code size of x86 debug builds, because X86 fast isel doesn't handle
arguments passed in memory at all. It only handles the x86_64 case of up
to 6 basic register parameters.
This is implemented by analyzing the entry block before ISel to identify
copy elision candidates. A copy elision candidate is an argument that is
used to fully initialize an alloca before any other possibly escaping
uses of that alloca. If an argument is a copy elision candidate, we set
a flag on the InputArg. If the the target generates loads from a fixed
stack object that matches the size and alignment requirements of the
alloca, the SelectionDAG builder will delete the stack object created
for the alloca and replace it with the fixed stack object. The load is
left behind to satisfy any remaining uses of the argument value. The
store is now dead and is therefore elided. The fixed stack object is
also marked as mutable, as it may now be modified by the user, and it
would be invalid to rematerialize the initial load from it.
Supersedes D28388
Fixes PR26328
Reviewers: chandlerc, MatzeB, qcolombet, inglorion, hans
Subscribers: igorb, llvm-commits
Differential Revision: https://reviews.llvm.org/D29668
llvm-svn: 296683
Resubmit r295336 after the bug with non-zero offset patterns on BE targets is fixed (r296336).
Support {a|s}ext, {a|z|s}ext load nodes as a part of load combine patters.
Reviewed By: filcab
Differential Revision: https://reviews.llvm.org/D29591
llvm-svn: 296651
Lower i1, i8 and i16 call parameters by extending them before storing them on
the stack. Also make sure we encode the correct, extended size in the
corresponding memory operand, and that we compute the correct stack size in the
end.
The latter is a bit more complicated because we used to compute the stack size
in the getStackAddress method, based on the Size and Offset of the parameters.
However, if the last parameter is sign extended, we'd be using the wrong,
non-extended size, and we'd end up with a smaller stack than we need to hold the
extended value. Instead of hacking this up based on the value of Size in
getStackAddress, we move our stack size handling logic to assignArg, where we
have access to the CCState which knows everything we could possibly want to know
about the stack. This way we don't need to duplicate any knowledge or resort to
any ugly hacks.
On this same occasion, update the IRTranslator test to check the sizes of the
stores everywhere, not just for sign extended paramteres.
llvm-svn: 296631
Recommiting after fixup of 32-bit aliasing sign offset bug in DAGCombiner.
* Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search and chain alias analysis which only
checks for parallel stores through the chain subgraph. This is cleaner
as the separation of non-interfering loads/stores from the
store-merging logic.
When merging stores search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited.
This improves the quality of the output SelectionDAG and the output
Codegen (save perhaps for some ARM cases where we correctly constructs
wider loads, but then promotes them to float operations which appear
but requires more expensive constant generation).
Some minor peephole optimizations to deal with improved SubDAG shapes (listed below)
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the chain aggregation in the merged stores across code
paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seems sufficient to not cause regressions in
tests.
5. Remove Chain dependencies of Memory operations on CopyfromReg
nodes as these are captured by data dependence
6. Forward loads-store values through tokenfactors containing
{CopyToReg,CopyFromReg} Values.
7. Peephole to convert buildvector of extract_vector_elt to
extract_subvector if possible (see
CodeGen/AArch64/store-merge.ll)
8. Store merging for the ARM target is restricted to 32-bit as
some in some contexts invalid 64-bit operations are being
generated. This can be removed once appropriate checks are
added.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable, improving load-store forwarding. One test in
particular is worth noting:
CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store
forwarding converts a load-store pair into a parallel store and
a memory-realized bitcast of the same value. However, because we
lose the sharing of the explicit and implicit store values we
must create another local store. A similar transformation
happens before SelectionDAG as well.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
llvm-svn: 296476
Lower i32, float and double parameters that need to live on the stack. This
boils down to creating some G_GEPs starting from the stack pointer and storing
the values there. During the process we also keep track of the stack size and
use the final value in the ADJCALLSTACKDOWN/UP instructions.
We currently assert for smaller types, since they usually require extensions.
They will be handled in a separate patch.
llvm-svn: 296473
Splitting critical edges when one of the source edges is an indirectbr
is hard in general (because it requires changing the memory the indirectbr
reads). But if a block only has a single indirectbr predecessor (which is
the common case), we can simulate splitting that edge by splitting
the destination block, and retargeting the *direct* branches.
This is motivated by the use of computed gotos in python 2.7: PyEval_EvalFrame()
ends up using an indirect branch with ~100 successors, and passing a constant to
each of those. Since MachineSink can't break indirect critical edges on demand
(and doing this in MIR doesn't look feasible), this causes us to emit about ~100
defs of registers containing constants, which we in the predecessor block, where
only one of those constants is used in each successor. So, at each computed goto,
we needlessly spill about a 100 constants to stack. The end result is that a
clang-compiled python interpreter can be about ~2.5x slower on a simple python
reduction loop than a gcc-compiled interpreter.
Differential Revision: https://reviews.llvm.org/D29916
llvm-svn: 296416
The transform in question claims to be doing:
// fold (add (select cc, 0, c), x) -> (select cc, x, (add, x, c))
...starting in PerformADDCombineWithOperands(), but it wasn't actually checking for a setcc node
for the sext/zext patterns.
This is exactly the opposite of a transform I'd like to add to DAGCombiner's foldSelectOfConstants(),
so I was seeing infinite loops with my draft of a patch applied.
The changes in select_const.ll look positive (less instructions). The change in arm-and-tst-peephole.ll
is unrelated. We're changing the input IR in that test to preserve the intent of the test, but that's
not affected by this code change.
Differential Revision:
https://reviews.llvm.org/D30355
llvm-svn: 296389
This pattern is essentially a i16 load from p+1 address:
%p1.i16 = bitcast i8* %p to i16*
%p2.i8 = getelementptr i8, i8* %p, i64 2
%v1 = load i16, i16* %p1.i16
%v2.i8 = load i8, i8* %p2.i8
%v2 = zext i8 %v2.i8 to i16
%v1.shl = shl i16 %v1, 8
%res = or i16 %v1.shl, %v2
Current implementation would identify %v1 load as the first byte load and would mistakenly emit a i16 load from %p1.i16 address. This patch adds a check that the first byte is loaded from a non-zero offset of the first load address. This way this address can be used as the base address for the combined value. Otherwise just give up combining.
llvm-svn: 296336
Recommiting after fixup of 32-bit aliasing sign offset bug in DAGCombiner.
* Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search and chain alias analysis which only
checks for parallel stores through the chain subgraph. This is cleaner
as the separation of non-interfering loads/stores from the
store-merging logic.
When merging stores search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited.
This improves the quality of the output SelectionDAG and the output
Codegen (save perhaps for some ARM cases where we correctly constructs
wider loads, but then promotes them to float operations which appear
but requires more expensive constant generation).
Some minor peephole optimizations to deal with improved SubDAG shapes (listed below)
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the chain aggregation in the merged stores across code
paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seems sufficient to not cause regressions in
tests.
5. Remove Chain dependencies of Memory operations on CopyfromReg
nodes as these are captured by data dependence
6. Forward loads-store values through tokenfactors containing
{CopyToReg,CopyFromReg} Values.
7. Peephole to convert buildvector of extract_vector_elt to
extract_subvector if possible (see
CodeGen/AArch64/store-merge.ll)
8. Store merging for the ARM target is restricted to 32-bit as
some in some contexts invalid 64-bit operations are being
generated. This can be removed once appropriate checks are
added.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable, improving load-store forwarding. One test in
particular is worth noting:
CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store
forwarding converts a load-store pair into a parallel store and
a memory-realized bitcast of the same value. However, because we
lose the sharing of the explicit and implicit store values we
must create another local store. A similar transformation
happens before SelectionDAG as well.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
llvm-svn: 296252
Splitting critical edges when one of the source edges is an indirectbr
is hard in general (because it requires changing the memory the indirectbr
reads). But if a block only has a single indirectbr predecessor (which is
the common case), we can simulate splitting that edge by splitting
the destination block, and retargeting the *direct* branches.
This is motivated by the use of computed gotos in python 2.7: PyEval_EvalFrame()
ends up using an indirect branch with ~100 successors, and passing a constant to
each of those. Since MachineSink can't break indirect critical edges on demand
(and doing this in MIR doesn't look feasible), this causes us to emit about ~100
defs of registers containing constants, which we in the predecessor block, where
only one of those constants is used in each successor. So, at each computed goto,
we needlessly spill about a 100 constants to stack. The end result is that a
clang-compiled python interpreter can be about ~2.5x slower on a simple python
reduction loop than a gcc-compiled interpreter.
Differential Revision: https://reviews.llvm.org/D29916
llvm-svn: 296149
The motivation for filling out these select-of-constants cases goes back to D24480,
where we discussed removing an IR fold from add(zext) --> select. And that goes back to:
https://reviews.llvm.org/rL75531https://reviews.llvm.org/rL159230
The idea is that we should always canonicalize patterns like this to a select-of-constants
in IR because that's the smallest IR and the best for value tracking. Note that we currently
do the opposite in some cases (like the cases in *this* patch). Ie, the proposed folds in
this patch already exist in InstCombine today:
https://github.com/llvm-mirror/llvm/blob/master/lib/Transforms/InstCombine/InstCombineSelect.cpp#L1151
As this patch shows, most targets generate better machine code for simple ext/add/not ops
rather than a select of constants. So the follow-up steps to make this less of a patchwork
of special-case folds and missing IR canonicalization:
1. Have DAGCombiner convert any select of constants into ext/add/not ops.
2 Have InstCombine canonicalize in the other direction (create more selects).
Differential Revision: https://reviews.llvm.org/D30180
llvm-svn: 296137
clang will generate IR like this for input using packed bitfields;
very simple semantically, but it's a bit tricky to actually
generate good code.
llvm-svn: 296080
Splitting critical edges when one of the source edges is an indirectbr
is hard in general (because it requires changing the memory the indirectbr
reads). But if a block only has a single indirectbr predecessor (which is
the common case), we can simulate splitting that edge by splitting
the destination block, and retargeting the *direct* branches.
This is motivated by the use of computed gotos in python 2.7: PyEval_EvalFrame()
ends up using an indirect branch with ~100 successors, and passing a constant to
each of those. Since MachineSink can't break indirect critical edges on demand
(and doing this in MIR doesn't look feasible), this causes us to emit about ~100
defs of registers containing constants, which we in the predecessor block, where
only one of those constants is used in each successor. So, at each computed goto,
we needlessly spill about a 100 constants to stack. The end result is that a
clang-compiled python interpreter can be about ~2.5x slower on a simple python
reduction loop than a gcc-compiled interpreter.
Differential Revision: https://reviews.llvm.org/D29916
llvm-svn: 296060
FastISel wasn't checking the isFPOnlySP subtarget feature before emitting
double-precision operations, so it got completely invalid CodeGen for doubles
on Cortex-M4F.
The normal ISel testing wasn't spectacular either so I added a second RUN line
to improve that while I was in the area.
llvm-svn: 296031
Introduce a common ValueHandler for call returns and formal arguments, and
inherit two different versions for handling the differences (at the moment the
only difference is the way physical registers are marked as used).
llvm-svn: 295973
Add support for lowering calls with parameters than can fit into regs. Use the
same ValueHandler that we used for function returns, but rename it to match its
new, extended purpose.
llvm-svn: 295971
The ARMConstantIslandPass didn't have support for handling accesses to
constant island objects through ARM::t2LDRBpci instructions. This adds
support for that.
This fixes PR31997.
llvm-svn: 295964
This patch adds missing sched classes for Thumb2 instructions.
This has been missing so far, and as a consequence, machine
scheduler models for individual sub-targets have tended to
be larger than they needed to be. These patches should help
write schedulers better and faster in the future
for ARM sub-targets.
Reviewer: Diana Picus
Differential Revision: https://reviews.llvm.org/D29953
llvm-svn: 295811
Summary:
This file was missed in the commit for Cortex-M23 and Cortex-M33
support. See https://reviews.llvm.org/D29073?id=85814 .
Reviewers: rengolin, javed.absar, samparker
Reviewed By: samparker
Subscribers: llvm-commits, aemerson
Differential Revision: https://reviews.llvm.org/D30162
llvm-svn: 295655
Start using the Subtarget to make decisions about what's legal. In particular,
we only mark floating point operations as legal if we have VFP2, which is
something we should've done from the very start.
llvm-svn: 295439
Resubmit -r295314 with PowerPC and AMDGPU tests updated.
Support {a|s}ext, {a|z|s}ext load nodes as a part of load combine patters.
Reviewed By: filcab
Differential Revision: https://reviews.llvm.org/D29591
llvm-svn: 295336
Support {a|s}ext, {a|z|s}ext load nodes as a part of load combine patters.
Reviewed By: filcab
Differential Revision: https://reviews.llvm.org/D29591
llvm-svn: 295314
Since they're only used for passing around double precision floating point
values into the general purpose registers, we'll lower them to VMOVDRR and
VMOVRRD.
llvm-svn: 295310
For now we just mark them as legal all the time and let the other passes bail
out if they can't handle it. In the future, we'll want to move more of the
brains into the legalizer.
llvm-svn: 295300
For the hard float calling convention, we just use the D registers.
For the soft-fp calling convention, we use the R registers and move values
to/from the D registers by means of G_SEQUENCE/G_EXTRACT. While doing so, we
make sure to honor the endianness of the target, since the CCAssignFn doesn't do
that for us.
For pure soft float targets, we still bail out because we don't support the
libcalls yet.
llvm-svn: 295295
Lay out trellis-shaped CFGs optimally.
A trellis of the shape below:
A B
|\ /|
| \ / |
| X |
| / \ |
|/ \|
C D
would be laid out A; B->C ; D by the current layout algorithm. Now we identify
trellises and lay them out either A->C; B->D or A->D; B->C. This scales with an
increasing number of predecessors. A trellis is a a group of 2 or more
predecessor blocks that all have the same successors.
because of this we can tail duplicate to extend existing trellises.
As an example consider the following CFG:
B D F H
/ \ / \ / \ / \
A---C---E---G---Ret
Where A,C,E,G are all small (Currently 2 instructions).
The CFG preserving layout is then A,B,C,D,E,F,G,H,Ret.
The current code will copy C into B, E into D and G into F and yield the layout
A,C,B(C),E,D(E),F(G),G,H,ret
define void @straight_test(i32 %tag) {
entry:
br label %test1
test1: ; A
%tagbit1 = and i32 %tag, 1
%tagbit1eq0 = icmp eq i32 %tagbit1, 0
br i1 %tagbit1eq0, label %test2, label %optional1
optional1: ; B
call void @a()
br label %test2
test2: ; C
%tagbit2 = and i32 %tag, 2
%tagbit2eq0 = icmp eq i32 %tagbit2, 0
br i1 %tagbit2eq0, label %test3, label %optional2
optional2: ; D
call void @b()
br label %test3
test3: ; E
%tagbit3 = and i32 %tag, 4
%tagbit3eq0 = icmp eq i32 %tagbit3, 0
br i1 %tagbit3eq0, label %test4, label %optional3
optional3: ; F
call void @c()
br label %test4
test4: ; G
%tagbit4 = and i32 %tag, 8
%tagbit4eq0 = icmp eq i32 %tagbit4, 0
br i1 %tagbit4eq0, label %exit, label %optional4
optional4: ; H
call void @d()
br label %exit
exit:
ret void
}
here is the layout after D27742:
straight_test: # @straight_test
; ... Prologue elided
; BB#0: # %entry ; A (merged with test1)
; ... More prologue elided
mr 30, 3
andi. 3, 30, 1
bc 12, 1, .LBB0_2
; BB#1: # %test2 ; C
rlwinm. 3, 30, 0, 30, 30
beq 0, .LBB0_3
b .LBB0_4
.LBB0_2: # %optional1 ; B (copy of C)
bl a
nop
rlwinm. 3, 30, 0, 30, 30
bne 0, .LBB0_4
.LBB0_3: # %test3 ; E
rlwinm. 3, 30, 0, 29, 29
beq 0, .LBB0_5
b .LBB0_6
.LBB0_4: # %optional2 ; D (copy of E)
bl b
nop
rlwinm. 3, 30, 0, 29, 29
bne 0, .LBB0_6
.LBB0_5: # %test4 ; G
rlwinm. 3, 30, 0, 28, 28
beq 0, .LBB0_8
b .LBB0_7
.LBB0_6: # %optional3 ; F (copy of G)
bl c
nop
rlwinm. 3, 30, 0, 28, 28
beq 0, .LBB0_8
.LBB0_7: # %optional4 ; H
bl d
nop
.LBB0_8: # %exit ; Ret
ld 30, 96(1) # 8-byte Folded Reload
addi 1, 1, 112
ld 0, 16(1)
mtlr 0
blr
The tail-duplication has produced some benefit, but it has also produced a
trellis which is not laid out optimally. With this patch, we improve the layouts
of such trellises, and decrease the cost calculation for tail-duplication
accordingly.
This patch produces the layout A,C,E,G,B,D,F,H,Ret. This layout does have
back edges, which is a negative, but it has a bigger compensating
positive, which is that it handles the case where there are long strings
of skipped blocks much better than the original layout. Both layouts
handle runs of executed blocks equally well. Branch prediction also
improves if there is any correlation between subsequent optional blocks.
Here is the resulting concrete layout:
straight_test: # @straight_test
; BB#0: # %entry ; A (merged with test1)
mr 30, 3
andi. 3, 30, 1
bc 12, 1, .LBB0_4
; BB#1: # %test2 ; C
rlwinm. 3, 30, 0, 30, 30
bne 0, .LBB0_5
.LBB0_2: # %test3 ; E
rlwinm. 3, 30, 0, 29, 29
bne 0, .LBB0_6
.LBB0_3: # %test4 ; G
rlwinm. 3, 30, 0, 28, 28
bne 0, .LBB0_7
b .LBB0_8
.LBB0_4: # %optional1 ; B (Copy of C)
bl a
nop
rlwinm. 3, 30, 0, 30, 30
beq 0, .LBB0_2
.LBB0_5: # %optional2 ; D (Copy of E)
bl b
nop
rlwinm. 3, 30, 0, 29, 29
beq 0, .LBB0_3
.LBB0_6: # %optional3 ; F (Copy of G)
bl c
nop
rlwinm. 3, 30, 0, 28, 28
beq 0, .LBB0_8
.LBB0_7: # %optional4 ; H
bl d
nop
.LBB0_8: # %exit
Differential Revision: https://reviews.llvm.org/D28522
llvm-svn: 295223
Summary:
Blocks ending in unreachable are typically cold because they end the
program or throw an exception, so merging them with other identical
blocks is usually profitable because it reduces the size of cold code.
MachineBlockPlacement generally does not arrange to fall through to such
blocks, so commoning these blocks will not introduce additional
unconditional branches.
Reviewers: hans, iteratee, haicheng
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D29153
llvm-svn: 295105
Backends don't support this yet. They would have to move to the swifterror
register before the tail call to make sure it is live-in to the call.
rdar://30495920
llvm-svn: 294982
Summary:
The attached test case fails with "fatal error: error in backend:
misaligned pc-relative fixup value" as the jump table is misaligned.
The EmitAlignment existed already for ARM and Thumb-1 code, but was
missing for Thumb-2.
The test checks that the fatal error disappears when generating an obj
file, as well as checking the align directive is there when producing an
asm file.
Reviewers: rengolin, grosbach, t.p.northover, jmolloy, SjoerdMeijer, samparker
Reviewed By: samparker
Subscribers: samparker, aemerson, llvm-commits
Differential Revision: https://reviews.llvm.org/D29650
llvm-svn: 294950
When generating a floating point comparison we currently unconditionally
generate VCMPE. This has the sideeffect of setting the cumulative Invalid
bit in FPSCR if any of the operands are QNaN.
It is expected that use of a relational predicate on a QNaN value should
raise Invalid. Quoting from the C standard:
The relational and equality operators support the usual mathematical
relationships between numeric values. For any ordered pair of numeric
values exactly one of relationships the less, greater, equal and is true.
Relational operators may raise the floating-point exception when argument
values are NaNs.
The standard doesn't explicitly state the expectation for equality operators,
but the implication and obvious expectation is that equality operators
should not raise Invalid on a QNaN input, as those predicates are wholly
defined on unordered inputs (to return not equal).
Therefore, add a new operand to ARMISD::FPCMP and FPCMPZ indicating if
QNaN should raise Invalid, and pipe that through to TableGen.
llvm-svn: 294945
In the encoding of system registers in the M-class MSR instruction the mask bits
should be 2 for registers that don't take a _<bits> qualifier (the instruction
is unpredictable otherwise), and should also be 2 if the register takes a
_<bits> qualifier but it's not present as no _<bits> is an alias for _nzcvq.
Differential Revision: https://reviews.llvm.org/D29828
llvm-svn: 294762
Gcc supports target armv7ve which is armv7-a with virtualization
extensions. This change adds support for this in llvm for gcc
compatibility.
Also remove redundant FeatureHWDiv, FeatureHWDivARM for a few models as
this is specified automatically by FeatureVirtualization.
Patch by Manoj Gupta.
Differential Revision: https://reviews.llvm.org/D29472
llvm-svn: 294661
If some of the trailing or leading bytes of a load combine pattern are zeroes we can combine the pattern to a load + zext and shift. Currently we don't support it, so the tests check the current codegen without load combine. This change will make the patch to support this kind of combine a bit more clear.
llvm-svn: 294591
Functions that have a dynamic alloca require a base register which is defined to
be X19 on AArch64 and r6 on ARM. We have defined the swifterror register to be
the same register. Use a different callee save register for swifterror instead:
X21 on AArch64
R8 on ARM
rdar://30433803
llvm-svn: 294551
We mark X0 as preserved by a call that passes the returned parameter.
x0 = ...
fun(x0) // no implicit def of x0
This no longer is valid if we pass the parameter in a different register then
the returned value as is the case with a swiftself parameter (passed in x20).
x20 = ...
fun(x20) // there should be an implict def of x8
rdar://30425845
llvm-svn: 294527
I forgot to remove the neonfp target feature from the test, which means we'd
have trouble selecting VADDS on targets that have neonfp enabled by default.
llvm-svn: 294451
Add a register bank for floating point values and select simple instructions
using them (add, copies from GPR).
This assumes that the hardware can cope with a single precision add (VADDS)
instruction, so the legalizer will treat G_FADD as legal and the instruction
selector will refuse to select if the hardware doesn't support it. In the future
we'll want to be more careful about this, and legalize to libcalls if we have to
use soft float.
llvm-svn: 294442
Currently we don't support these nodes, so the tests check the current codegen without load combine. This change makes the review of the change to support these nodes more clear.
Separated from https://reviews.llvm.org/D29591 review.
llvm-svn: 294305
When constructing global address literals while targeting the RWPI
relocation model. LLVM currently only uses literal pools. If MOVW/MOVT
instructions are available we can use these instead. Beside being more
efficient it allows -arm-execute-only to work with
-relocation-model=RWPI as well.
When we generate MOVW/MOVT for global addresses when targeting the RWPI
relocation model, we need to use base relative relocations. This patch
does the needed plumbing in MC to generate these for MOVW/MOVT.
Differential Revision: https://reviews.llvm.org/D29487
Change-Id: I446786e43a6f5aa9b6a5bb2cd216d60d41c7755d
llvm-svn: 294298
Summary:
The tail call optimisation is performed before register allocation, so
at that point we don't know if LR is being spilt or not. If LR was spilt
to the stack, then we cannot do a tail call optimisation. That would
involve popping back into LR which is not possible in Thumb1 code.
Reviewers: rengolin, jmolloy, rovka, olista01
Reviewed By: olista01
Subscribers: llvm-commits, aemerson
Differential Revision: https://reviews.llvm.org/D29020
llvm-svn: 294000
Summary:
llc would hit a fatal error for errors in inline assembly. The
diagnostics message is now printed.
Reviewers: rengolin, MatzeB, javed.absar, anemet
Reviewed By: anemet
Subscribers: jyknight, nemanjai, llvm-commits
Differential Revision: https://reviews.llvm.org/D29408
llvm-svn: 293999
This is the second in the series of patches to enable adding
of machine sched-models for ARM processors easier and compact.
This patch focuses on integer instructions and adds missing
sched definitions.
Reviewers: rovka, rengolin
Differential Revision: https://reviews.llvm.org/D29127
llvm-svn: 293935
Recommiting after fixing X86 inc/dec chain bug.
* Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search and chain alias analysis which only
checks for parallel stores through the chain subgraph. This is cleaner
as the separation of non-interfering loads/stores from the
store-merging logic.
When merging stores search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited.
This improves the quality of the output SelectionDAG and the output
Codegen (save perhaps for some ARM cases where we correctly constructs
wider loads, but then promotes them to float operations which appear
but requires more expensive constant generation).
Some minor peephole optimizations to deal with improved SubDAG shapes (listed below)
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the chain aggregation in the merged stores across code
paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seems sufficient to not cause regressions in
tests.
5. Remove Chain dependencies of Memory operations on CopyfromReg
nodes as these are captured by data dependence
6. Forward loads-store values through tokenfactors containing
{CopyToReg,CopyFromReg} Values.
7. Peephole to convert buildvector of extract_vector_elt to
extract_subvector if possible (see
CodeGen/AArch64/store-merge.ll)
8. Store merging for the ARM target is restricted to 32-bit as
some in some contexts invalid 64-bit operations are being
generated. This can be removed once appropriate checks are
added.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable, improving load-store forwarding. One test in
particular is worth noting:
CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store
forwarding converts a load-store pair into a parallel store and
a memory-realized bitcast of the same value. However, because we
lose the sharing of the explicit and implicit store values we
must create another local store. A similar transformation
happens before SelectionDAG as well.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
llvm-svn: 293893
It is important to change the ArgInfo's type from pointer to integer, otherwise
the CC assign function won't know what to do. Instead of hacking it up, we use
ComputeValueVTs and introduce some of the helpers that we will need later on for
lowering more complex types.
llvm-svn: 293889
Add both cores to the target parser and TableGen. Test that eabi
attributes are set correctly for both cores. Additionally, test the
absence and presence of MOVT in Cortex-M23 and Cortex-M33, respectively.
Committed on behalf of Sanne Wouda.
Reviewers : rengolin, olista01.
Differential Revision: https://reviews.llvm.org/D29073
llvm-svn: 293761
When choosing the best successor for a block, ordinarily we would have preferred
a block that preserves the CFG unless there is a strong probability the other
direction. For small blocks that can be duplicated we now skip that requirement
as well, subject to some simple frequency calculations.
Differential Revision: https://reviews.llvm.org/D28583
llvm-svn: 293716
The Requires class overrides the target requirements of an instruction,
rather than adding to them, so all ARM instructions need to include the
IsARM predicate when they have overwitten requirements.
This caused the swp and swpb instructions to be allowed in thumb mode
assembly, and the ARM encoding of CDP to be selected in codegen (which
is different for conditional instructions).
Differential Revision: https://reviews.llvm.org/D29283
llvm-svn: 293634
Support lowering AEABI TLS access (__aeabi_read_tp) with long calls.
This requires adjusting the call sequence to use an indirect call to get
full addressability.
Resolves PR31769!
llvm-svn: 293433
The interleaved access pass is an IR-to-IR transformation that runs before code
generation. It matches interleaved memory operations to target-specific
intrinsics (that are later lowered to load and store multiple instructions on
ARM/AArch64). We place tests for similar passes (e.g., GlobalMergePass) under
test/Transforms. This patch moves the InterleavedAccessPass tests out of
test/CodeGen and into target-specific directories under
test/Transforms/InterleavedAccess.
Although the pass is an IR pass, many of the existing tests were llc tests
rather opt tests. For example, the tests would check for ldN/stN instructions
generated by llc rather than the intrinsic calls the pass actually inserts.
Thus, this patch updates all tests to be opt tests that check for the inserted
intrinsics. We already have separate CodeGen tests that ensure we lower the
interleaved access intrinsics to their corresponding ldN/stN instructions. In
addition to migrating the tests to opt, this patch also performs some minor
clean-up (to ensure consistent naming, etc.).
Differential Revision: https://reviews.llvm.org/D29184
llvm-svn: 293309
The Windows on ARM target uses custom division for normal division as
the backend needs to insert division-by-zero checks. However, it is
designed to only handle non-vectorized division. ARM has custom
lowering for vectorized division as that can avoid loading registers
with the values and invoke a division routine for each one, preferring
to lower using NEON instructions. Fall back to the custom lowering for
the NEON instructions if we encounter a vectorized division.
Resolves PR31778!
llvm-svn: 293259
* Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search and chain alias analysis which only
checks for parallel stores through the chain subgraph. This is cleaner
as the separation of non-interfering loads/stores from the
store-merging logic.
When merging stores search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited.
This improves the quality of the output SelectionDAG and the output
Codegen (save perhaps for some ARM cases where we correctly constructs
wider loads, but then promotes them to float operations which appear
but requires more expensive constant generation).
Some minor peephole optimizations to deal with improved SubDAG shapes (listed below)
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the chain aggregation in the merged stores across code
paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seems sufficient to not cause regressions in
tests.
5. Remove Chain dependencies of Memory operations on CopyfromReg
nodes as these are captured by data dependence
6. Forward loads-store values through tokenfactors containing
{CopyToReg,CopyFromReg} Values.
7. Peephole to convert buildvector of extract_vector_elt to
extract_subvector if possible (see
CodeGen/AArch64/store-merge.ll)
8. Store merging for the ARM target is restricted to 32-bit as
some in some contexts invalid 64-bit operations are being
generated. This can be removed once appropriate checks are
added.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable, improving load-store forwarding. One test in
particular is worth noting:
CodeGen/PowerPC/ppc64-align-long-double.ll - Improved load-store
forwarding converts a load-store pair into a parallel store and
a memory-realized bitcast of the same value. However, because we
lose the sharing of the explicit and implicit store values we
must create another local store. A similar transformation
happens before SelectionDAG as well.
Reviewers: arsenm, hfinkel, tstellarAMD, jyknight, nhaehnle
llvm-svn: 293184
Add support for loading i1, i8 and i16 arguments from the stack, with or without
the ABI extension flags.
When the ABI extension flags are present, we load a 4-byte value, otherwise we
preserve the size of the load and let the instruction selector replace it with a
LDRB/LDRH. This generates the same thing as DAGISel.
Differential Revision: https://reviews.llvm.org/D27803
llvm-svn: 293163
Later code expects the vector loads produced to be directly
concatenable, which means we shouldn't pad anything except the last load
produced with UNDEF.
llvm-svn: 293088
The previous patch (https://reviews.llvm.org/rL289538) got reverted because of a bug. Chandler also requested some changes to the algorithm.
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20161212/413479.html
This is an updated patch. The key difference is that collectBitProviders (renamed to calculateByteProvider) now collects the origin of one byte, not the whole value. It simplifies the implementation and allows to stop the traversal earlier if we know that the result won't be used.
From the original commit:
Match a pattern where a wide type scalar value is loaded by several narrow loads and combined by shifts and ors. Fold it into a single load or a load and a bswap if the targets supports it.
Assuming little endian target:
i8 *a = ...
i32 val = a[0] | (a[1] << 8) | (a[2] << 16) | (a[3] << 24)
=>
i32 val = *((i32)a)
i8 *a = ...
i32 val = (a[0] << 24) | (a[1] << 16) | (a[2] << 8) | a[3]
=>
i32 val = BSWAP(*((i32)a))
This optimization was discussed on llvm-dev some time ago in "Load combine pass" thread. We came to the conclusion that we want to do this transformation late in the pipeline because in presence of atomic loads load widening is irreversible transformation and it might hinder other optimizations.
Eventually we'd like to support folding patterns like this where the offset has a variable and a constant part:
i32 val = a[i] | (a[i + 1] << 8) | (a[i + 2] << 16) | (a[i + 3] << 24)
Matching the pattern above is easier at SelectionDAG level since address reassociation has already happened and the fact that the loads are adjacent is clear. Understanding that these loads are adjacent at IR level would have involved looking through geps/zexts/adds while looking at the addresses.
The general scheme is to match OR expressions by recursively calculating the origin of individual bytes which constitute the resulting OR value. If all the OR bytes come from memory verify that they are adjacent and match with little or big endian encoding of a wider value. If so and the load of the wider type (and bswap if needed) is allowed by the target generate a load and a bswap if needed.
Reviewed By: RKSimon, filcab, chandlerc
Differential Revision: https://reviews.llvm.org/D27861
llvm-svn: 293036
Add support for:
* i1 add
* i1 function arguments, if passed through registers
* i1 returns, with ABI signext/zeroext
Differential Revision: https://reviews.llvm.org/D27706
llvm-svn: 293035
At the moment, this means supporting the signext/zeroext attribute on the return
type of the function. For function arguments, signext/zeroext should be handled
by the caller, so there's nothing for us to do until we start lowering calls.
Note that this does not include support for other extensions (i8 to i16), those
will be added later.
Differential Revision: https://reviews.llvm.org/D27705
llvm-svn: 293034
This is a series of patches to enable adding of machine sched
models for ARM processors easier and compact. They define new
sched-readwrites for groups of ARM instructions. This has been
missing so far, and as a consequence, machine scheduler models
for individual sub-targets have tended to be larger than they
needed to be.
The current patch focuses on floating-point instructions.
Reviewers: Diana Picus (rovka), Renato Golin (rengolin)
Differential Revision: https://reviews.llvm.org/D28194
llvm-svn: 292825
We also want to optimise tests like this: return a*b == 0. The MULS
instruction is flag setting, so we don't need the CMP instruction but can
instead branch on the result of the MULS. The generated instructions sequence
for this example was: MULS, MOVS, MOVS, CMP. The MOVS instruction load the
boolean values resulting from the select instruction, but these MOVS
instructions are flag setting and were thus preventing this optimisation. Now
we first reorder and move the MULS to before the CMP and generate sequence
MOVS, MOVS, MULS, CMP so that the optimisation could trigger. Reordering of the
MULS and MOVS is safe to do because the subsequent MOVS instructions just set
the CPSR register and don't use it, i.e. the CPSR is dead.
Differential Revision: https://reviews.llvm.org/D27990
llvm-svn: 292608
Diana Picus