1. LiveIntervals now implement a 4 slot per instruction model. Load,
Use, Def and a Store slot. This is required in order to correctly
represent caller saved register clobbering on function calls,
register reuse in the same instruction (def resues last use) and
also spill code added later by the allocator. The previous
representation (2 slots per instruction) was insufficient and as a
result was causing subtle bugs.
2. Fixes in spill code generation. This was the major cause of
failures in the test suite.
3. Linear scan now has core support for folding memory operands. This
is untested and not enabled (the live interval update function does
not attempt to fold loads/stores in instructions).
4. Lots of improvements in the debugging output of both live intervals
and linear scan. Give it a try... it is beautiful :-)
In summary the above fixes all the issues with the recent reserved
register elimination changes and get the allocator very close to the
next big step: folding memory operands.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11654 91177308-0d34-0410-b5e6-96231b3b80d8
by operator<< on MachineInstr's, and looking up what register "24" is all of the
time was greatly annoying.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11623 91177308-0d34-0410-b5e6-96231b3b80d8
that need them. This is very useful on CISCy targets like the X86 because it
reduces the total spill pressure, and makes better use of it's (large)
instruction set. Though the X86 backend doesn't know how to rewrite many
instructions yet, this already makes a substantial difference on 176.gcc for
example:
Before:
Time:
8.0099 ( 31.2%) 0.0100 ( 12.5%) 8.0199 ( 31.2%) 7.7186 ( 30.0%) Local Register Allocator
Code quality:
734559 asm-printer - Number of machine instrs printed
111395 ra-local - Number of registers reloaded
79902 ra-local - Number of registers spilled
231554 x86-peephole - Number of peephole optimization performed
After:
Time:
7.8700 ( 30.6%) 0.0099 ( 19.9%) 7.8800 ( 30.6%) 7.7892 ( 30.2%) Local Register Allocator
Code quality:
733083 asm-printer - Number of machine instrs printed
2379 ra-local - Number of reloads fused into instructions
109046 ra-local - Number of registers reloaded
79881 ra-local - Number of registers spilled
230658 x86-peephole - Number of peephole optimization performed
So by fusing 2300 instructions, we reduced the static number of instructions
by 1500, and reduces the number of peepholes (and thus the work) by about 900.
This also clearly reduces the number of reload/spill instructions that are
emitted.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11542 91177308-0d34-0410-b5e6-96231b3b80d8
prototypes, even if they don't precisely match what it would prefer to use.
This fixes: CBackend/2004-02-15-PreexistingExternals.llx compiling it into:
ltmp_0_30 = memcpy(l14_C, 4u, 17);
ltmp_1_30 = memcpy(((int *)l27_A), ((unsigned )(long)l27_B), ((int )123u));
instead of:
ltmp_0_30 = memcpy(l14_C, 4u, 17);
ltmp_1_27 = l43_memcpy(l27_A, l27_B, 123u);
Which does the wrong thing as you could imagine.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11481 91177308-0d34-0410-b5e6-96231b3b80d8
MRegisterInfo::getNumRegs() instead of
MRegisterInfo::FirstVirtualRegister.
Also use MRegisterInfo::is{Physical,Virtual}Register where
appropriate.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11477 91177308-0d34-0410-b5e6-96231b3b80d8
that will be responsible for the creation of MachineFunctions and will
be required by all MachineFunctionPass passes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11453 91177308-0d34-0410-b5e6-96231b3b80d8
Rename SetMachineOperandConst's formal parameters to match other methods here.
Mark some methods as being used only by the SPARC back-end.
Fix a missing-paren bug in OutputValue().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11363 91177308-0d34-0410-b5e6-96231b3b80d8
MachineBasicBlock. Also change opcode to a short and numImplicitRefs
to an unsigned char so that overall MachineInstr's size stays the
same.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11357 91177308-0d34-0410-b5e6-96231b3b80d8
ilist of MachineInstr objects. This allows constant time removal and
insertion of MachineInstr instances from anywhere in each
MachineBasicBlock. It also allows for constant time splicing of
MachineInstrs into or out of MachineBasicBlocks.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11340 91177308-0d34-0410-b5e6-96231b3b80d8
instead of randomly groping about inside its outEdges array.
Make SchedGraph::addDummyEdges() use getNumOutEdges() instead of
outEdges.size().
Get rid of ifdefed-out code in SchedGraph::buildGraph().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11238 91177308-0d34-0410-b5e6-96231b3b80d8
the Virt2PhysRegMap std::map with an std::vector. This speeds up the
register allocator another (almost) 40%, from .72->.45s in a release build
of LLC on 253.perlbmk.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11219 91177308-0d34-0410-b5e6-96231b3b80d8
This speeds up live variables a lot, from .60/.39s -> .47/.26s in LLC, for
the first/second pass respectively.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11216 91177308-0d34-0410-b5e6-96231b3b80d8
from physical registers, and they are always dense, it makes sense to not have
a ton of RBtree overhead. This change speeds up regalloclocal about ~30% on
253.perlbmk, from .35s -> .27s in the JIT (in LLC, it goes from .74 -> .55).
Now live variable analysis is the slowest codegen pass. Of course it doesn't
help that we have to run it twice, because regalloclocal doesn't update it,
but even if it did it would be the slowest pass (now it's just the 2x slowest
pass :(
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11215 91177308-0d34-0410-b5e6-96231b3b80d8
slots each. As a concequence they get numbered as 0, 2, 4 and so
on. The first slot is used for operand uses and the second for
defs. Here's an example:
0: A = ...
2: B = ...
4: C = A + B ;; last use of A
The live intervals should look like:
A = [1, 5)
B = [3, x)
C = [5, y)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11141 91177308-0d34-0410-b5e6-96231b3b80d8
registers (not as the max number of registers).
Change toSpill from a std::set into a std::vector<bool>.
Use the reverse iterator adapter to do a reverse scan of allocatable
registers.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11061 91177308-0d34-0410-b5e6-96231b3b80d8
Move Passes.h (which defines the interface to this file) to the top.
Move statistics to the top of the file.
Add a comment
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11034 91177308-0d34-0410-b5e6-96231b3b80d8
of a linear search to find the first range for comparisons. This cuts
down the linear scan register allocator running time by a factor of 3
in 254.perlbmk and by a factor of 2.2 in 176.gcc.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11030 91177308-0d34-0410-b5e6-96231b3b80d8
Simplification of LiveIntervals::Interval::overlaps() and addition of
examples to overlaps() and liveAt() to make them clearer.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@11028 91177308-0d34-0410-b5e6-96231b3b80d8
when joining we need to check if we overlap with the second interval
or any of its aliases.
Also make joining intervals the default.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10973 91177308-0d34-0410-b5e6-96231b3b80d8
is a move between two registers, at least one of the registers is
virtual and the two live intervals do not overlap.
This results in about 40% reduction in intervals, 30% decrease in the
register allocators running time and a 20% increase in peephole
optimizations (mainly move eliminations).
The option can be enabled by passing -join-liveintervals where
appropriate.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10965 91177308-0d34-0410-b5e6-96231b3b80d8
virtReg lives on the stack. Now a virtual register has an entry in the
virtual->physical map or the virtual->stack slot map but never in
both.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10958 91177308-0d34-0410-b5e6-96231b3b80d8
LiveIntervals::Interval::liveAt. Both were considering the live ranges
closed in the end, when they are actually open.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10835 91177308-0d34-0410-b5e6-96231b3b80d8
when an implicitely defined register is later used by an alias. For example:
call foo
%reg1024 = mov %AL
The call implicitely defines EAX but only AL is used. Before this fix
no information was available on AL. Now EAX and all its aliases except
AL get defined and die at the call instruction whereas AL lives to be
killed by the assignment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10813 91177308-0d34-0410-b5e6-96231b3b80d8
LiveVariables::HandlePhysRegDef private they use information that is
not in memory when LiveVariables finishes the analysis.
Also update the TwoAddressInstructionPass to not use this interface.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10755 91177308-0d34-0410-b5e6-96231b3b80d8
This should get hunked over to the Sparc backend, along with
MachineCodeForInstruction and a bunch of files in include/llvm/Codegen,
but those battles will have to wait for a later time.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10731 91177308-0d34-0410-b5e6-96231b3b80d8
of the register allocator as follows:
before after
mesa 2.3790 1.5994
vpr 2.6008 1.2078
gcc 1.9840 0.5273
mcf 0.2569 0.0470
eon 1.8468 1.4359
twolf 0.9475 0.2004
burg 1.6807 1.3300
lambda 1.2191 0.3764
Speedups range anyware from 30% to over 400% :-)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10712 91177308-0d34-0410-b5e6-96231b3b80d8
A = B op C where A == C, but this cannot really occur in practice
because of SSA form. Add an assert to check that just to be safe.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10682 91177308-0d34-0410-b5e6-96231b3b80d8
saved register it has a longer free range than ECX (which is defined
every time there is a fnuction call) which makes ECX a better register
to reserve.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10635 91177308-0d34-0410-b5e6-96231b3b80d8
which denotes the register we would like to be assigned to (virtual or
physical). In register allocation, if this hint exists and we can map
it to a physical register (it is either a physical register or it is a
virtual register that already got assigned to a physical one) we use
that register if it is available instead of a random one in the free
pool.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10634 91177308-0d34-0410-b5e6-96231b3b80d8
with live intervals was missing registers that were used before they
were defined (in the arbitrary order live intervals numbers
instructions).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10603 91177308-0d34-0410-b5e6-96231b3b80d8
nesting level when computing it. Right now the allocator uses:
w = sum_over_defs_uses( 10 ^ nesting level );
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10569 91177308-0d34-0410-b5e6-96231b3b80d8
* Move sparc specific code out of generic code
* Eliminate the getOffset() method which made INVALID_FRAME_OFFSET
necessary, which made pulling in MAX_INT as a sentinal necessary.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10553 91177308-0d34-0410-b5e6-96231b3b80d8
instruction pass. This also fixes all remaining bugs for this new
allocator to pass all tests under test/Programs.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10515 91177308-0d34-0410-b5e6-96231b3b80d8
more operands and the two first operands are constrained to be the
same. The pass takes an instruction of the form:
a = b op c
and transforms it into:
a = b
a = a op c
and also preserves live variables.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10512 91177308-0d34-0410-b5e6-96231b3b80d8
killing instruction is tracked. This causes the LiveIntervals to
create bogus intervals. The workaound is to add a range to the
interval from the redefinition to the end of the basic block.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10510 91177308-0d34-0410-b5e6-96231b3b80d8
Move some of the longer LiveIntervals::Interval method out of the
header and add debug information to them. Fix bug and simplify range
merging code.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10509 91177308-0d34-0410-b5e6-96231b3b80d8
a) remove opIsUse(), opIsDefOnly(), opIsDefAndUse()
b) add isUse(), isDef()
c) rename opHiBits32() to isHiBits32(),
opLoBits32() to isLoBits32(),
opHiBits64() to isHiBits64(),
opLoBits64() to isLoBits64().
This results to much more readable code, for example compare
"op.opIsDef() || op.opIsDefAndUse()" to "op.isDef()" a pattern used
very often in the code.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10461 91177308-0d34-0410-b5e6-96231b3b80d8
bug where spill instructions were added to the next basic block
instead of the end of the current one if the instruction that required
the spill was the last in the block.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10272 91177308-0d34-0410-b5e6-96231b3b80d8
externally-visible linkage, and SaveStateToModule must default to true for llc.
I don't remember why I made it const; perhaps it should be deconstified.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9858 91177308-0d34-0410-b5e6-96231b3b80d8
each instruction produces as "operand" -1, and the other operands as 0
.. n, as before. PhyRegAlloc::saveState() is refactored into
PhyRegAlloc::saveStateForValue().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9842 91177308-0d34-0410-b5e6-96231b3b80d8
implementing verifySavedState().
In saveState(), use the new AllocInfo::AllocStateTy enum, and increment
Insn each time through the loop.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9617 91177308-0d34-0410-b5e6-96231b3b80d8
Prototype option to save state in a global instead of as a Constant in
the Module. (Turned off, for now, with the on/off switch welded in the off
position. You get the idea.)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9500 91177308-0d34-0410-b5e6-96231b3b80d8
Make FnAllocState contain vectors of AllocInfo, instead of LLVM Constants.
Give doFinalization a method comment, and let it do the work of converting
AllocInfos to LLVM Constants.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9451 91177308-0d34-0410-b5e6-96231b3b80d8
* Fix order of #includes
* Make code layout more consistent
* Eliminate extraneous whitespace and comment-lines
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9433 91177308-0d34-0410-b5e6-96231b3b80d8
* Make file description more readable
* Make code layout more consistent, include comment in assert so it's visible
during execution if it hits
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9430 91177308-0d34-0410-b5e6-96231b3b80d8
Try to improve method comments a little.
Get rid of some excess whitespace; put braces on previous line when possible.
Add stub for method to verify the work of saveState().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@9385 91177308-0d34-0410-b5e6-96231b3b80d8
and TargetInstrDescriptor::ImplicitUses to always point to a null
terminated array and never be null. So there is no need to check for
pointer validity when iterating over those sets. Code that looked
like:
if (const unsigned* AS = TID.ImplicitDefs) {
for (int i = 0; AS[i]; ++i) {
// use AS[i]
}
}
was changed to:
for (const unsigned* AS = TID.ImplicitDefs; *AS; ++AS) {
// use *AS
}
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@8960 91177308-0d34-0410-b5e6-96231b3b80d8
Shorten the markSuggestedColorUsable method.
Add a switch for saving reg. alloc. state (coming soon).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@8697 91177308-0d34-0410-b5e6-96231b3b80d8
(I also zapped printMachineCode() and printLabel() at the previous checkin,
but forgot to mention it.)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@8646 91177308-0d34-0410-b5e6-96231b3b80d8
PhyRegAlloc.cpp:
Don't include TargetMachine.h or TargetRegInfo.h, because these are provided
by PhyRegAlloc.h.
Merge class RegisterAllocator into class PhyRegAlloc.
Simplify & move ctor, dtor to PhyRegAlloc.h.
Make some of PhyRegAlloc's reference members into pointer members,
so they can be more easily messed with.
MarkAllocatedRegs() becomes a member method, with fewer args.
PhyRegAlloc.h:
Include Pass.h, TargetMachine.h and TargetRegInfo.h. Don't declare
TargetRegInfo forward.
Give AddedInstrns the obvious clear() method.
Make some of PhyRegAlloc's reference members into pointer members,
so they can be more easily messed with.
Add prototype for markAllocatedRegs().
Remove unused inline void constructLiveRanges().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@8641 91177308-0d34-0410-b5e6-96231b3b80d8