On Cortex-A8, we use the NEON v2f32 instructions for f32 arithmetic. For
better latency, we also send D-register copies down the NEON pipeline by
translating them to vorr instructions.
This patch promotes even S-register copies to D-register copies when
possible so they can also go down the NEON pipeline. Example:
vldr.32 s0, LCPI0_0
loop:
vorr d1, d0, d0
loop2:
...
vadd.f32 d1, d1, d16
The vorr instruction looked like this after regalloc:
%S2<def> = COPY %S0, %D1<imp-def>
Copies involving odd S-registers, and copies that don't define the full
D-register are left alone.
llvm-svn: 137182
The tSpill and tRestore instructions are just copies of the tSTRspi and
tLDRspi instructions, respectively. Just use those directly instead.
llvm-svn: 134092
sink them into MC layer.
- Added MCInstrInfo, which captures the tablegen generated static data. Chang
TargetInstrInfo so it's based off MCInstrInfo.
llvm-svn: 134021
is, it assumes addresses are 64-bit aligned (which should be the more common
case). If the alignment is found not to be aligned, then getOperandLatency()
would adjust the operand latency computation by one to compensate for it.
rdar://9294833
llvm-svn: 129742
entries being compared may not be ARMConstantPoolValue. Without checking
whether they are ARMConstantPoolValue first, and if the stars and moons
are aligned properly, the equality test may return true (when the first few
words of two Constants' values happen to be identical) and very bad things can
happen.
rdar://9125354
llvm-svn: 128203
int tries = INT_MAX;
while (tries > 0) {
tries--;
}
The check should be:
subs r4, #1
cmp r4, #0
bgt LBB0_1
The subs can set the overflow V bit when r4 is INT_MAX+1 (which loop
canonicalization apparently does in this case). cmp #0 would have cleared
it while not changing the N and Z bits. Since BGT is dependent on the V
bit, i.e. (N == V) && !Z, it is not safe to eliminate the cmp #0.
rdar://9172742
llvm-svn: 128179
1. Fixed ARM pc adjustment.
2. Fixed dynamic-no-pic codegen
3. CSE of pc-relative load of global addresses.
It's now enabled by default for Darwin.
llvm-svn: 123991
flags. They are still not enable in this revision.
Added TargetInstrInfo::isZeroCost() to fix a fundamental problem with
the scheduler's model of operand latency in the selection DAG.
Generalized unit tests to work with sched-cycles.
llvm-svn: 123969
value, the "add pc" must be CSE'ed at the same time. We could follow the same
approach as T2 by adding pseudo instructions that combine the ldr + "add pc".
But the better approach is to use movw + movt (which I will enable soon), so
I'll leave this as a TODO.
llvm-svn: 123949
TargetInstrInfo:
Change produceSameValue() to take MachineRegisterInfo as an optional argument.
When in SSA form, targets can use it to make more aggressive equality analysis.
Machine LICM:
1. Eliminate isLoadFromConstantMemory, use MI.isInvariantLoad instead.
2. Fix a bug which prevent CSE of instructions which are not re-materializable.
3. Use improved form of produceSameValue.
ARM:
1. Teach ARM produceSameValue to look pass some PIC labels.
2. Look for operands from different loads of different constant pool entries
which have same values.
3. Re-implement PIC GA materialization using movw + movt. Combine the pair with
a "add pc" or "ldr [pc]" to form pseudo instructions. This makes it possible
to re-materialize the instruction, allow machine LICM to hoist the set of
instructions out of the loop and make it possible to CSE them. It's a bit
hacky, but it significantly improve code quality.
4. Some minor bug fixes as well.
With the fixes, using movw + movt to materialize GAs significantly outperform the
load from constantpool method. 186.crafty and 255.vortex improved > 20%, 254.gap
and 176.gcc ~10%.
llvm-svn: 123905
movw r0, :lower16:(L_foo$non_lazy_ptr-(LPC0_0+4))
movt r0, :upper16:(L_foo$non_lazy_ptr-(LPC0_0+4))
LPC0_0:
add r0, pc, r0
It's not yet enabled by default as some tests are failing. I suspect bugs in
down stream tools.
llvm-svn: 123619
DAG scheduling during isel. Most new functionality is currently
guarded by -enable-sched-cycles and -enable-sched-hazard.
Added InstrItineraryData::IssueWidth field, currently derived from
ARM itineraries, but could be initialized differently on other targets.
Added ScheduleHazardRecognizer::MaxLookAhead to indicate whether it is
active, and if so how many cycles of state it holds.
Added SchedulingPriorityQueue::HasReadyFilter to allowing gating entry
into the scheduler's available queue.
ScoreboardHazardRecognizer now accesses the ScheduleDAG in order to
get information about it's SUnits, provides RecedeCycle for bottom-up
scheduling, correctly computes scoreboard depth, tracks IssueCount, and
considers potential stall cycles when checking for hazards.
ScheduleDAGRRList now models machine cycles and hazards (under
flags). It tracks MinAvailableCycle, drives the hazard recognizer and
priority queue's ready filter, manages a new PendingQueue, properly
accounts for stall cycles, etc.
llvm-svn: 122541
Use the same COPY_TO_REGCLASS approach as for the 2-register *_sfp instructions.
This change made a big difference in the code generated for the
CodeGen/Thumb2/cross-rc-coalescing-2.ll test: The coalescer is still doing
a fine job, but some instructions that were previously moved outside the loop
are not moved now. It's using fewer VFP registers now, which is generally
a good thing, so I think the estimates for register pressure changed and that
affected the LICM behavior. Since that isn't obviously wrong, I've just
changed the test file. This completes the work for Radar 8711675.
llvm-svn: 121730
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
additional pipeline stall. So it's frequently better to single codegen
vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
vmla + vmla is very bad. But this isn't ideal either:
vmul
vadd
vmla
Instead, we want to expand the second vmla:
vmla
vmul
vadd
Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
faster.
Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.
A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
vmla / vmls will trigger one of the special hazards.
Work in progress, only A+B are enabled.
llvm-svn: 120960
Remove movePastCSLoadStoreOps and associated code for simple pointer
increments. Update routines that depended upon other opcodes for save/restore.
Adjust all testcases accordingly.
llvm-svn: 119725
and xor. The 32-bit move immediates can be hoisted out of loops by machine
LICM but the isel hacks were preventing them.
Instead, let peephole optimization pass recognize registers that are defined by
immediates and the ARM target hook will fold the immediates in.
Other changes include 1) do not fold and / xor into cmp to isel TST / TEQ
instructions if there are multiple uses. This happens when the 'and' is live
out, machine sink would have sinked the computation and that ends up pessimizing
code. The peephole pass would recognize situations where the 'and' can be
toggled to define CPSR and eliminate the comparison anyway.
2) Move peephole pass to after machine LICM, sink, and CSE to avoid blocking
important optimizations.
rdar://8663787, rdar://8241368
llvm-svn: 119548
'db', 'ib', 'da') instead of having that mode as a separate field in the
instruction. It's more convenient for the asm parser and much more readable for
humans.
<rdar://problem/8654088>
llvm-svn: 119310
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
llvm-svn: 118135
Chad Rosier