1. Fix pre-ra scheduler so it doesn't try to push instructions above calls to
"optimize for latency". Call instructions don't have the right latency and
this is more likely to use introduce spills.
2. Fix if-converter cost function. For ARM, it should use instruction latencies,
not # of micro-ops since multi-latency instructions is completely executed
even when the predicate is false. Also, some instruction will be "slower"
when they are predicated due to the register def becoming implicit input.
rdar://8598427
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118135 91177308-0d34-0410-b5e6-96231b3b80d8
There were a number of issues to fix up here:
* The "device" argument of the llvm.memory.barrier intrinsic should be
used to distinguish the "Full System" domain from the "Inner Shareable"
domain. It has nothing to do with using DMB vs. DSB instructions.
* The compiler should never need to emit DSB instructions. Remove the
ARMISD::SYNCBARRIER node and also remove the instruction patterns for DSB.
* Merge the separate DMB/DSB instructions for options only used for the
disassembler with the default DMB/DSB instructions. Add the default
"full system" option ARM_MB::SY to the ARM_MB::MemBOpt enum.
* Add a separate ARMISD::MEMBARRIER_MCR node for subtargets that implement
a data memory barrier using the MCR instruction.
* Fix up encodings for these instructions (except MCR).
I also updated the tests and added a few new ones to check for DMB options
that were not currently being exercised.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117756 91177308-0d34-0410-b5e6-96231b3b80d8
operand and one of them has a single use that is a live out copy, favor the
one that is live out. Otherwise it will be difficult to eliminate the copy
if the instruction is a loop induction variable update. e.g.
BB:
sub r1, r3, #1
str r0, [r2, r3]
mov r3, r1
cmp
bne BB
=>
BB:
str r0, [r2, r3]
sub r3, r3, #1
cmp
bne BB
This fixed the recent 256.bzip2 regression.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117675 91177308-0d34-0410-b5e6-96231b3b80d8
- Initial register pressure in the loop should be all the live defs into the
loop. Not just those from loop preheader which is often empty.
- When an instruction is hoisted, update register pressure from loop preheader
to the original BB.
- Treat only use of a virtual register as kill since the code is still SSA.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116956 91177308-0d34-0410-b5e6-96231b3b80d8
"long latency" enough to hoist even if it may increase spilling. Reloading
a value from spill slot is often cheaper than performing an expensive
computation in the loop. For X86, that means machine LICM will hoist
SQRT, DIV, etc. ARM will be somewhat aggressive with VFP and NEON
instructions.
- Enable register pressure aware machine LICM by default.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116781 91177308-0d34-0410-b5e6-96231b3b80d8
callee-saved registers at the end of the lists. Also prefer to avoid using
the low registers that are in register subclasses required by certain
instructions, so that those registers will more likely be available when needed.
This change makes a huge improvement in spilling in some cases. Thanks to
Jakob for helping me realize the problem.
Most of this patch is fixing the testsuite. There are quite a few places
where we're checking for specific registers. I changed those to wildcards
in places where that doesn't weaken the tests. The spill-q.ll and
thumb2-spill-q.ll tests stopped spilling with this change, so I added a bunch
of live values to force spills on those tests.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116055 91177308-0d34-0410-b5e6-96231b3b80d8
between the high and low registers for prologue/epilogue code. This was
a Darwin-only thing that wasn't providing a realistic benefit anymore.
Combining the save areas simplifies the compiler code and results in better
ARM/Thumb2 codegen.
For example, previously we would generate code like:
push {r4, r5, r6, r7, lr}
add r7, sp, #12
stmdb sp!, {r8, r10, r11}
With this change, we combine the register saves and generate:
push {r4, r5, r6, r7, r8, r10, r11, lr}
add r7, sp, #12
rdar://8445635
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114340 91177308-0d34-0410-b5e6-96231b3b80d8
take multiple cycles to decode.
For the current if-converter clients (actually only ARM), the instructions that
are predicated on false are not nops. They would still take machine cycles to
decode. Micro-coded instructions such as LDM / STM can potentially take multiple
cycles to decode. If-converter should take treat them as non-micro-coded
simple instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113570 91177308-0d34-0410-b5e6-96231b3b80d8
the VST pseudos. The VLD/VST scheduling still needs work (see pr6722), but
at least we shouldn't confuse the loads with the stores.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113473 91177308-0d34-0410-b5e6-96231b3b80d8
"For ARM stack frames that utilize variable sized objects and have either
large local stack areas or require dynamic stack realignment, allocate a
base register via which to access the local frame. This allows efficient
access to frame indices not accessible via the FP (either due to being out
of range or due to dynamic realignment) or the SP (due to variable sized
object allocation). In particular, this greatly improves efficiency of access
to spill slots in Thumb functions which contain VLAs."
r112986 fixed a latent bug exposed by the above.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112989 91177308-0d34-0410-b5e6-96231b3b80d8
large local stack areas or require dynamic stack realignment, allocate a
base register via which to access the local frame. This allows efficient
access to frame indices not accessible via the FP (either due to being out
of range or due to dynamic realignment) or the SP (due to variable sized
object allocation). In particular, this greatly improves efficiency of access
to spill slots in Thumb functions which contain VLAs.
rdar://7352504
rdar://8374540
rdar://8355680
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112883 91177308-0d34-0410-b5e6-96231b3b80d8
int x(int t) {
if (t & 256)
return -26;
return 0;
}
We generate this:
tst.w r0, #256
mvn r0, #25
it eq
moveq r0, #0
while gcc generates this:
ands r0, r0, #256
it ne
mvnne r0, #25
bx lr
Scandalous really!
During ISel time, we can look for this particular pattern. One where we have a
"MOVCC" that uses the flag off of a CMPZ that itself is comparing an AND
instruction to 0. Something like this (greatly simplified):
%r0 = ISD::AND ...
ARMISD::CMPZ %r0, 0 @ sets [CPSR]
%r0 = ARMISD::MOVCC 0, -26 @ reads [CPSR]
All we have to do is convert the "ISD::AND" into an "ARM::ANDS" that sets [CPSR]
when it's zero. The zero value will all ready be in the %r0 register and we only
need to change it if the AND wasn't zero. Easy!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112664 91177308-0d34-0410-b5e6-96231b3b80d8
Update all the tests using those intrinsics and add support for
auto-upgrading bitcode files with the old versions of the intrinsics.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112271 91177308-0d34-0410-b5e6-96231b3b80d8
comparison that would overflow.
- The other under/overflow cases can't actually happen because the immediates
which would trigger them are legal (so we don't enter this code), but
adjusted the style to make it clear the transform is always valid.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112053 91177308-0d34-0410-b5e6-96231b3b80d8
printing "lsl #0". This fixes the remaining parts of pr7792. Make
corresponding changes for encoding/decoding these instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111251 91177308-0d34-0410-b5e6-96231b3b80d8
platform. It's apparently "bl __muldf3" on linux, for example. Since that's
not what we're checking here, it's more robust to just force a triple. We
just wwant to check that the inline FP instructions are only generated
on cpus that have them."
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110830 91177308-0d34-0410-b5e6-96231b3b80d8
memory and synchronization barrier dmb and dsb instructions.
- Change instruction names to something more sensible (matching name of actual
instructions).
- Added tests for memory barrier codegen.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110785 91177308-0d34-0410-b5e6-96231b3b80d8
Also added a test case to check for the added benefit of this patch: it's optimizing away the unnecessary restore of sp from fp for some non-leaf functions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110707 91177308-0d34-0410-b5e6-96231b3b80d8
have 4 bits per register in the operand encoding), but have undefined
behavior when the operand value is 13 or 15 (SP and PC, respectively).
The trivial coalescer in linear scan sometimes will merge a copy from
SP into a subsequent instruction which uses the copy, and if that
instruction cannot legally reference SP, we get bad code such as:
mls r0,r9,r0,sp
instead of:
mov r2, sp
mls r0, r9, r0, r2
This patch adds a new register class for use by Thumb2 that excludes
the problematic registers (SP and PC) and is used instead of GPR
for those operands which cannot legally reference PC or SP. The
trivial coalescer explicitly requires that the register class
of the destination for the COPY instruction contain the source
register for the COPY to be considered for coalescing. This prevents
errant instructions like that above.
PR7499
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109842 91177308-0d34-0410-b5e6-96231b3b80d8
integers with mov + vdup. 8003375. This is
currently disabled by default because LICM will
not hoist a VDUP, so it pessimizes the code if
the construct occurs inside a loop (8248029).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@109799 91177308-0d34-0410-b5e6-96231b3b80d8
instruction for non-constant operands. This includes the case referenced
in the README.txt regarding a bitfield copy.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108608 91177308-0d34-0410-b5e6-96231b3b80d8
and a combine pattern to use it for setting a bit-field to a constant
value. More to come for non-constant stores.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108570 91177308-0d34-0410-b5e6-96231b3b80d8
in the literal field of an instruction. E.g.,
long long foo(long long a) {
return a - 734439407618LL;
}
rdar://7038284
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108339 91177308-0d34-0410-b5e6-96231b3b80d8
Add explicit testcases for tail calls within the same module.
Duplicate some code to humor those who think .w doesn't apply on ARM.
Leave this disabled on Thumb1, and add some comments explaining why it's hard
and won't gain much.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@107851 91177308-0d34-0410-b5e6-96231b3b80d8
There are 2 changes relative to the previous version of the patch:
1) For the "simple" if-conversion case, there's no need to worry about
RemoveExtraEdges not handling an unanalyzable branch. Predicated terminators
are ignored in this context, so RemoveExtraEdges does the right thing.
This might break someday if we ever treat indirect branches (BRIND) as
predicable, but for now, I just removed this part of the patch, because
in the case where we do not add an unconditional branch, we rely on keeping
the fall-through edge to CvtBBI (which is empty after this transformation).
The change relative to the previous patch is:
@@ -1036,10 +1036,6 @@
IterIfcvt = false;
}
- // RemoveExtraEdges won't work if the block has an unanalyzable branch,
- // which is typically the case for IfConvertSimple, so explicitly remove
- // CvtBBI as a successor.
- BBI.BB->removeSuccessor(CvtBBI->BB);
RemoveExtraEdges(BBI);
// Update block info. BB can be iteratively if-converted.
2) My patch exposed a bug in the code for merging the tail of a "diamond",
which had previously never been exercised. The code was simply checking that
the tail had a single predecessor, but there was a case in
MultiSource/Benchmarks/VersaBench/dbms where that single predecessor was
neither edge of the diamond. I added the following change to check for
that:
@@ -1276,7 +1276,18 @@
// tail, add a unconditional branch to it.
if (TailBB) {
BBInfo TailBBI = BBAnalysis[TailBB->getNumber()];
- if (TailBB->pred_size() == 1 && !TailBBI.HasFallThrough) {
+ bool CanMergeTail = !TailBBI.HasFallThrough;
+ // There may still be a fall-through edge from BBI1 or BBI2 to TailBB;
+ // check if there are any other predecessors besides those.
+ unsigned NumPreds = TailBB->pred_size();
+ if (NumPreds > 1)
+ CanMergeTail = false;
+ else if (NumPreds == 1 && CanMergeTail) {
+ MachineBasicBlock::pred_iterator PI = TailBB->pred_begin();
+ if (*PI != BBI1->BB && *PI != BBI2->BB)
+ CanMergeTail = false;
+ }
+ if (CanMergeTail) {
MergeBlocks(BBI, TailBBI);
TailBBI.IsDone = true;
} else {
With these fixes, I was able to run all the SingleSource and MultiSource
tests successfully.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@107110 91177308-0d34-0410-b5e6-96231b3b80d8
regressions.
--- Reverse-merging r106939 into '.':
U test/CodeGen/Thumb2/thumb2-ifcvt3.ll
U lib/CodeGen/IfConversion.cpp
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106951 91177308-0d34-0410-b5e6-96231b3b80d8
if-conversion. The RemoveExtraEdges function doesn't work for blocks that
end with unanalyzable branches, so in those cases, the "extra" edges must
be explicitly removed. The CopyAndPredicateBlock and MergeBlocks methods
can also avoid copying successor edges due to branches that have already
been removed. The latter case is especially helpful when MergeBlocks is
called for handling "diamond" if-conversions, where otherwise you can end
up with some weird intermediate states in the CFG. Unfortunately I've
been unable to find cases where this cleanup actually makes a significant
difference in the code. There is one test where we manage to remove an
empty block at the end of a function. Radar 6911268.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106939 91177308-0d34-0410-b5e6-96231b3b80d8
