In particular, this preserves the kill flag, which allows the Thumb2 cbn?z
optimization to be applied in cases where a branch has been re-created after
the live variables analysis pass, e.g. by the machine block placement pass.
This appears to be low risk; a number of other targets seem to already be
doing something similar, e.g. AArch64, PowerPC.
Differential Revision: http://reviews.llvm.org/D9184
llvm-svn: 235639
This makes it more likely that we can use the 16-bit push and pop instructions
on Thumb-2, saving around 4 bytes per function.
Differential Revision: http://reviews.llvm.org/D9165
llvm-svn: 235637
This appears to have been introduced back in r76698 as part of an unrelated
change. I can find no official ARM documentation stating that Thumb-2 functions
require 4-byte alignment; in fact, ARM documentation appears to contradict
this (see, e.g., ARM Architecture Reference Manual Thumb-2 Supplement,
section 2.6.1: "Thumb-2 enforces 16-bit alignment on all instructions.").
Also remove code that sets alignment for ARM functions, which is redundant
with code in the MachineFunction constructor, and remove the hidden
-arm-align-constant-islands flag, which has been enabled by default since
r146739 (Dec 2011) and has probably received sufficient testing by now.
Differential Revision: http://reviews.llvm.org/D9138
llvm-svn: 235636
Third time's the charm. The previous commit was reverted as a
reverse for-loop in SelectionDAGBuilder::lowerWorkItem did 'I--'
on an iterator at the beginning of a vector, causing asserts
when using debugging iterators. This commit fixes that.
llvm-svn: 235608
This is a re-commit of r235101, which also fixes the problems with the previous patch:
- Switches with only a default case and non-fallthrough were handled incorrectly
- The previous patch tickled a bug in PowerPC Early-Return Creation which is fixed here.
> This is a major rewrite of the SelectionDAG switch lowering. The previous code
> would lower switches as a binary tre, discovering clusters of cases
> suitable for lowering by jump tables or bit tests as it went along. To increase
> the likelihood of finding jump tables, the binary tree pivot was selected to
> maximize case density on both sides of the pivot.
>
> By not selecting the pivot in the middle, the binary trees would not always
> be balanced, leading to performance problems in the generated code.
>
> This patch rewrites the lowering to search for clusters of cases
> suitable for jump tables or bit tests first, and then builds the binary
> tree around those clusters. This way, the binary tree will always be balanced.
>
> This has the added benefit of decoupling the different aspects of the lowering:
> tree building and jump table or bit tests finding are now easier to tweak
> separately.
>
> For example, this will enable us to balance the tree based on profile info
> in the future.
>
> The algorithm for finding jump tables is quadratic, whereas the previous algorithm
> was O(n log n) for common cases, and quadratic only in the worst-case. This
> doesn't seem to be major problem in practice, e.g. compiling a file consisting
> of a 10k-case switch was only 30% slower, and such large switches should be rare
> in practice. Compiling e.g. gcc.c showed no compile-time difference. If this
> does turn out to be a problem, we could limit the search space of the algorithm.
>
> This commit also disables all optimizations during switch lowering in -O0.
>
> Differential Revision: http://reviews.llvm.org/D8649
llvm-svn: 235560
Summary:
In the f16-promote test, make the checks for native conversion instructions
similar to the libcall checks:
- Remove hard coded register names
- Do not check exact instruction sequences.
This fixes test flakiness due to non-determinism in instruction
scheduling and register allocation. I also fixed a few minor things in
the CHECK-LIBCALL checks.
I'll try to find a way to check that unnecessary loads, stores, or
conversions don't happen.
Reviewers: mzolotukhin, srhines, ab
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D9112
llvm-svn: 235363
Instead of merging everything together, look at the users of
GlobalVariables, and try to group them by function, to create
sets of globals used "together".
Using that information, a less-aggressive alternative is to keep merging
everything together *except* globals that are only ever used alone, that
is, those for which it's clearly non-profitable to merge with others.
In my testing, grouping by Function is too aggressive, but grouping by
BasicBlock is too conservative. Anything in-between isn't trivially
available, so stick with Function grouping for now.
cl::opts are added for testing; both enabled by default.
A few of the testcases aren't testing the merging proper, but just
various edge cases when merging does occur. Update them to use the
previous grouping behavior. Also, one of the tests is unrelated to
GlobalMerge; change it accordingly.
While there, switch to r234666' flags rather than the brutal -O3.
Differential Revision: http://reviews.llvm.org/D8070
llvm-svn: 235249
Summary:
This patch adds legalization support to operate on FP16 as a load/store type
and do operations on it as floats.
Tests for ARM are added to test/CodeGen/ARM/fp16-promote.ll
Reviewers: srhines, t.p.northover
Differential Revision: http://reviews.llvm.org/D8755
llvm-svn: 235215
See r230786 and r230794 for similar changes to gep and load
respectively.
Call is a bit different because it often doesn't have a single explicit
type - usually the type is deduced from the arguments, and just the
return type is explicit. In those cases there's no need to change the
IR.
When that's not the case, the IR usually contains the pointer type of
the first operand - but since typed pointers are going away, that
representation is insufficient so I'm just stripping the "pointerness"
of the explicit type away.
This does make the IR a bit weird - it /sort of/ reads like the type of
the first operand: "call void () %x(" but %x is actually of type "void
()*" and will eventually be just of type "ptr". But this seems not too
bad and I don't think it would benefit from repeating the type
("void (), void () * %x(" and then eventually "void (), ptr %x(") as has
been done with gep and load.
This also has a side benefit: since the explicit type is no longer a
pointer, there's no ambiguity between an explicit type and a function
that returns a function pointer. Previously this case needed an explicit
type (eg: a function returning a void() function was written as
"call void () () * @x(" rather than "call void () * @x(" because of the
ambiguity between a function returning a pointer to a void() function
and a function returning void).
No ambiguity means even function pointer return types can just be
written alone, without writing the whole function's type.
This leaves /only/ the varargs case where the explicit type is required.
Given the special type syntax in call instructions, the regex-fu used
for migration was a bit more involved in its own unique way (as every
one of these is) so here it is. Use it in conjunction with the apply.sh
script and associated find/xargs commands I've provided in rr230786 to
migrate your out of tree tests. Do let me know if any of this doesn't
cover your cases & we can iterate on a more general script/regexes to
help others with out of tree tests.
About 9 test cases couldn't be automatically migrated - half of those
were functions returning function pointers, where I just had to manually
delete the function argument types now that we didn't need an explicit
function type there. The other half were typedefs of function types used
in calls - just had to manually drop the * from those.
import fileinput
import sys
import re
pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)')
addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$")
func_end = re.compile("(?:void.*|\)\s*)\*$")
def conv(match, line):
if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)):
return line
return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():]
for line in sys.stdin:
sys.stdout.write(conv(re.search(pat, line), line))
llvm-svn: 235145
This is a major rewrite of the SelectionDAG switch lowering. The previous code
would lower switches as a binary tre, discovering clusters of cases
suitable for lowering by jump tables or bit tests as it went along. To increase
the likelihood of finding jump tables, the binary tree pivot was selected to
maximize case density on both sides of the pivot.
By not selecting the pivot in the middle, the binary trees would not always
be balanced, leading to performance problems in the generated code.
This patch rewrites the lowering to search for clusters of cases
suitable for jump tables or bit tests first, and then builds the binary
tree around those clusters. This way, the binary tree will always be balanced.
This has the added benefit of decoupling the different aspects of the lowering:
tree building and jump table or bit tests finding are now easier to tweak
separately.
For example, this will enable us to balance the tree based on profile info
in the future.
The algorithm for finding jump tables is O(n^2), whereas the previous algorithm
was O(n log n) for common cases, and quadratic only in the worst-case. This
doesn't seem to be major problem in practice, e.g. compiling a file consisting
of a 10k-case switch was only 30% slower, and such large switches should be rare
in practice. Compiling e.g. gcc.c showed no compile-time difference. If this
does turn out to be a problem, we could limit the search space of the algorithm.
This commit also disables all optimizations during switch lowering in -O0.
Differential Revision: http://reviews.llvm.org/D8649
llvm-svn: 235101
Remove 'inlinedAt:' from MDLocalVariable. Besides saving some memory
(variables with it seem to be single largest `Metadata` contributer to
memory usage right now in -g -flto builds), this stops optimization and
backend passes from having to change local variables.
The 'inlinedAt:' field was used by the backend in two ways:
1. To tell the backend whether and into what a variable was inlined.
2. To create a unique id for each inlined variable.
Instead, rely on the 'inlinedAt:' field of the intrinsic's `!dbg`
attachment, and change the DWARF backend to use a typedef called
`InlinedVariable` which is `std::pair<MDLocalVariable*, MDLocation*>`.
This `DebugLoc` is already passed reliably through the backend (as
verified by r234021).
This commit removes the check from r234021, but I added a new check
(that will survive) in r235048, and changed the `DIBuilder` API in
r235041 to require a `!dbg` attachment whose 'scope:` is in the same
`MDSubprogram` as the variable's.
If this breaks your out-of-tree testcases, perhaps the script I used
(mdlocalvariable-drop-inlinedat.sh) will help; I'll attach it to PR22778
in a moment.
llvm-svn: 235050
Add missing `!dbg` attachments to `@llvm.dbg.*` intrinsics. I updated
these using a script (add-dbg-to-intrinsics.sh) that I'll attach to
PR22778 for posterity.
llvm-svn: 235040
Many of these predate llvm-readobj. With elf-dump we had to match
a relocation to symbol number and symbol number to symbol name or
section number.
llvm-svn: 235015
This commit makes LLVM not estimate branch probabilities when doing a
single bit bitmask tests.
The code that originally made me discover this is:
if ((a & 0x1) == 0x1) {
..
}
In this case we don't actually have any branch probability information
and should not assume to have any. LLVM transforms this into:
%and = and i32 %a, 1
%tobool = icmp eq i32 %and, 0
So, in this case, the result of a bitwise and is compared against 0,
but nevertheless, we should not assume to have probability
information.
CodeGen/ARM/2013-10-11-select-stalls.ll started failing because the
changed probabilities changed the results of
ARMBaseInstrInfo::isProfitableToIfCvt() and led to an Ifcvt of the
diamond in the test. AFAICT, the test was never meant to test this and
thus changing the test input slightly to not change the probabilities
seems like the best way to preserve the meaning of the test.
llvm-svn: 234979
Fill in the TODO in CodeGenPrepare::OptimizeCallInst so that global
variables that are passed to memory intrinsics are aligned in the same
way that allocas are.
Differential Revision: http://reviews.llvm.org/D8421
llvm-svn: 234735
Currently, there's a single flag, checked by the pass itself.
It can't force-enable the pass (and is on by default), because it
might not even have been created, as that's the targets decision.
Instead, have separate explicit flags, so that the decision is
consistently made in the target.
Keep the flag as a last-resort "force-disable GlobalMerge" for now,
for backwards compatibility.
llvm-svn: 234666
Currently, llvm (backend) doesn't know cortex-r4, even though it is the
default target for armv7r. Using "--target=armv7r-arm-none-eabi" provokes
'cortex-r4' is not a recognized processor for this target' by llvm.
This patch adds support for cortex-r4 and, very closely related, r4f.
llvm-svn: 234486
Because -menable-no-nans causes fcmp conditions to be rewritten
without 'o' or 'u' the recognition code in needs to cope. Also
extended it to handle 'le' and 'ge.
Differential Revision: http://reviews.llvm.org/D8725
llvm-svn: 234421
Summary: Looks like new code from [[ http://reviews.llvm.org/rL222057 | rL222057 ]] doesn't account for early `return` in `ARMFrameLowering::emitPrologue`, which leads to loosing `.cfi_def_cfa_offset` directive for functions without stack frame.
Reviewers: echristo, rengolin, asl, t.p.northover
Reviewed By: t.p.northover
Subscribers: llvm-commits, rengolin, aemerson
Differential Revision: http://reviews.llvm.org/D8606
llvm-svn: 234399
Check that the `MDLocalVariable::getInlinedAt()` in a debug info
intrinsic's variable always matches the `MDLocation::getInlinedAt()` of
its `!dbg` attachment.
The goal here is to get rid of `MDLocalVariable::getInlinedAt()`
entirely (PR22778), since it's expensive and unnecessary, but I'll let
this verifier check bake for a while (a week maybe?) first. I've
updated the testcases that had the wrong value for `inlinedAt:`.
This checks that things are sane in the IR, but currently things go out
of whack in a few places in the backend. I'll follow shortly with
assertions in the backend (with code fixes).
If you have out-of-tree testcases that just started failing, here's how
I updated these ones:
1. The verifier check gives you the basic block, function, instruction,
and relevant metadata arguments (metadata numbering doesn't
necessarily match the source file, unfortunately).
2. Look at the `@llvm.dbg.*()` instruction, and compare the
`inlinedAt:` fields of the variable argument (second `metadata`
argument) and the `!dbg` attachment.
3. Figure out based on the variable `scope:` chain and the functions in
the file whether the variable has been inlined (and into what), so
you can determine which `inlinedAt:` is actually correct. In all of
the in-tree testcases, the `!MDLocation()` was correct and the
`!MDLocalVariable()` was wrong, but YMMV.
4. Duplicate the metadata that you're going to change, and add/drop the
`inlinedAt:` field from one of them. Be careful that the other
references to the same metadata node point at the correct one.
llvm-svn: 234021
Register coalescing can change the target of a RegPair hint to a
physreg, we should not crash on this. This also slightly improved the
way ARMBaseRegisterInfo::updateRegAllocHint() works.
llvm-svn: 233987
it more liberally.
SplitVecOp_TRUNCATE has logic for recursively splitting oversize vectors
that need more than one round of splitting to become legal. There are many
other ISD nodes that could benefit from this logic, so factor it out and
use it for FP_TO_UINT,FP_TO_SINT,SINT_TO_FP,UINT_TO_FP and FTRUNC.
llvm-svn: 233681
Fix debug info in these tests, which started failing with a WIP patch to
verify compile units and types. The problems look like they were all
caused by bitrot. They fell into these categories:
- Using `!{i32 0}` instead of `!{}`.
- Using `!{null}` instead of `!{}`.
- Using `!MDExpression()` instead of `!{}`.
- Using `!8` instead of `!{!8}`.
- `file:` references that pointed at `MDCompileUnit`s instead of the
same `MDFile` as the compile unit.
- `file:` references that were numerically off-by-one or (off-by-ten).
llvm-svn: 233415
"Fix the MachineScheduler's logic for updating ready times for in-order.
Now the scheduler updates a node's ready time as soon as it is
scheduled, before releasing dependent nodes."
This fix was only made in one variant of the ScheduleDAGMI driver.
Francois de Ferriere reported the issue in the other bit of code where
it was also needed.
I never got around to coming up with a test case, but it's an
obvious fix that shouldn't be delayed any longer.
I'll try to refactor this code a little better.
I did verify performance on a wide variety of targets and saw no
negative impact with this fix.
llvm-svn: 233366
Summary:
The ARM backend can use a loop to implement copying byval parameters before
a call. In non-thumb2 mode it uses a constant pool load to materialize the
trip count. For targets that need movt instead (e.g. Native Client), use
the same code as in thumb2 mode to materialize the trip count.
Reviewers: jfb, t.p.northover
Differential Revision: http://reviews.llvm.org/D8442
llvm-svn: 233324
The pass used to be enabled by default with CodeGenOpt::Less (-O1).
This is too aggressive, considering the pass indiscriminately merges
all globals together.
Currently, performance doesn't always improve, and, on code that uses
few globals (e.g., the odd file- or function- static), more often than
not is degraded by the optimization. Lengthy discussion can be found
on llvmdev (AArch64-focused; ARM has similar problems):
http://lists.cs.uiuc.edu/pipermail/llvmdev/2015-February/082800.html
Also, it makes tooling and debuggers less useful when dealing with
globals and data sections.
GlobalMerge needs to better identify those cases that benefit, and this
will be done separately. In the meantime, move the pass to run with
-O3 rather than -O1, on both ARM and AArch64.
llvm-svn: 233024
This change is incorrect since it converts double rounding into single rounding,
which can produce different results. Instead this optimization will be done by
modifying Clang's codegen to not produce double rounding in the first place.
This reverts commit r232954.
llvm-svn: 232962
Specifically when the conversion is done in two steps, f16 -> f32 -> f64.
For example:
%1 = tail call float @llvm.convert.from.fp16.f32(i16 %0)
%conv = fpext float %1 to double
to:
vcvtb.f64.f16
llvm-svn: 232954
If we couldn't analyze its terminator (i.e., it's an indirectbr, or some
other weirdness), we can't safely re-if-convert a predicated block,
because we can't tell whether the predicated terminator can
fallthrough (it does).
Currently, we would completely ignore the fallthrough successor. In
the added testcase, this means we used to generate:
...
@ %entry:
cmp r5, #21
ittt ne
@ %cc1f:
cmpne r7, #42
@ %cc2t:
strne.w r5, [r8]
movne pc, r10
@ %cc1t:
...
Whereas the successor of %cc1f was originally %bb1.
With the fix, we get the correct:
...
@ %entry:
cmp r5, #21
itt eq
@ %cc1t:
streq.w r5, [r11]
moveq pc, r0
@ %cc1f:
cmp r7, #42
itt ne
@ %cc2t:
strne.w r5, [r8]
movne pc, r10
@ %bb1:
...
rdar://20192768
Differential Revision: http://reviews.llvm.org/D8509
llvm-svn: 232872
The code this patch removes was there to make sure the text sections went
before the dwarf sections. That is necessary because MachO uses offsets
relative to the start of the file, so adding a section can change relaxations.
The dwarf sections were being printed at the start just to produce symbols
pointing at the start of those sections.
The underlying issue was fixed in r231898. The dwarf sections are now printed
when they are about to be used, which is after we printed the text sections.
To make sure we don't regress, the patch makes the MachO streamer assert
if CodeGen puts anything unexpected after the DWARF sections.
llvm-svn: 232842
This is very related to the bug fixed in r174431. The problem is that
SelectionDAG does not include alignment in the uniquing of loads and
stores. When an otherwise no-op DAGCombine would increase the alignment
of a load or store, the original node would be returned (with the
alignment increased), which would cause the node not to be processed by
any further DAGCombines.
I don't have a direct testcase for this that manifests on an in-tree
target, but I did see some noise in the tests for other targets and have
updated them for it.
llvm-svn: 232780
Memcpy, and other memory intrinsics, typically tries to use LDM/STM if
the source and target addresses are 4-byte aligned. In CodeGenPrepare
look for calls to memory intrinsics and, if the object is on the
stack, 4-byte align it if it's large enough that we expect that memcpy
would want to use LDM/STM to copy it.
Differential Revision: http://reviews.llvm.org/D7908
llvm-svn: 232627
COFF COMDATs (for selection kinds other than 'select any') require at
least one non-section symbol in the symbol table.
Satisfy this by morally enhancing the linkage from private to internal.
Differential Revision: http://reviews.llvm.org/D8394
llvm-svn: 232570
The input offset to needsFrameBaseReg is a negative value below the top of the
stack frame, but when converting to a positive offset from the bottom of the
stack frame this value was negated, causing the final offset to be too large
by twice the input offset's magnitude. Fix that by not negating the offset.
Patch by John Brawn
Differential Revision: http://reviews.llvm.org/D8316
llvm-svn: 232513
ARMv6K is another layer between ARMV6 and ARMV6T2. This is the LLVM
side of the changes.
ARMV6 family LLVM implementation.
+-------------------------------------+
| ARMV6 |
+----------------+--------------------+
| ARMV6M (thumb) | ARMV6K (arm,thumb) | <- From ARMV6K and ARMV6M processors
+----------------+--------------------+ have support for hint instructions
| ARMV6T2 (arm,thumb,thumb2) | (SEV/WFE/WFI/NOP/YIELD). They can
+-------------------------------------+ be either real or default to NOP.
| ARMV7 (arm,thumb,thumb2) | The two processors also use
+-------------------------------------+ different encoding for them.
Patch by Vinicius Tinti.
llvm-svn: 232468
Similar to gep (r230786) and load (r230794) changes.
Similar migration script can be used to update test cases, which
successfully migrated all of LLVM and Polly, but about 4 test cases
needed manually changes in Clang.
(this script will read the contents of stdin and massage it into stdout
- wrap it in the 'apply.sh' script shown in previous commits + xargs to
apply it over a large set of test cases)
import fileinput
import sys
import re
rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL)
def conv(match):
line = match.group(1)
line += match.group(4)
line += ", "
line += match.group(2)
return line
line = sys.stdin.read()
off = 0
for match in re.finditer(rep, line):
sys.stdout.write(line[off:match.start()])
sys.stdout.write(conv(match))
off = match.end()
sys.stdout.write(line[off:])
llvm-svn: 232184
The main issue being fixed here is that APCS targets handling a "byval align N"
parameter with N > 4 were miscounting what objects were where on the stack,
leading to FrameLowering setting the frame pointer incorrectly and clobbering
the stack.
But byval handling had grown over many years, and had multiple layers of cruft
trying to compensate for each other and calculate padding correctly. This only
really needs to be done once, in the HandleByVal function. Elsewhere should
just do what it's told by that call.
I also stripped out unnecessary APCS/AAPCS distinctions (now that Clang emits
byvals with the correct C ABI alignment), which simplified HandleByVal.
rdar://20095672
llvm-svn: 231959
to disable lane switching if we don't actually have the instruction
set we want to switch to. Models the earlier check above the
conditional for the pass.
The testcase is one that triggered with the assert that's added
as part of the fix, use it to avoid adding a new testcase as it
highlights the same problem.
llvm-svn: 231539
This is based on the following equivalences:
select(C0 & C1, X, Y) <=> select(C0, select(C1, X, Y), Y)
select(C0 | C1, X, Y) <=> select(C0, X, select(C1, X, Y))
Many target cannot perform and/or on the CPU flags and therefore the
right side should be choosen to avoid materializign the i1 flags in an
integer register. If the target can perform this operation efficiently
we normalize to the left form.
Differential Revision: http://reviews.llvm.org/D7622
llvm-svn: 231507
This commit enables forming vector extloads for ARM.
It only does so for legal types, and when we can't fold the extension
in a wide/long form of the user instruction.
Enabling it for larger types isn't as good an idea on ARM as it is on
X86, because:
- we pretend that extloads are legal, but end up generating vld+vmov
- we have instructions like vld {dN, dM}, which can't be generated
when we "manually expand" extloads to vld+vmov.
For legal types, the combine doesn't fire that often: in the
integration tests only in a big endian testcase, where it removes a
pointless AND.
Related to rdar://19723053
Differential Revision: http://reviews.llvm.org/D7423
llvm-svn: 231396
