- Use SplitBlockPredecessors to factor out common predecessors of the critical edge destination. This is disabled for now due to some regressions.
llvm-svn: 61248
wrappers around the interesting code and use an obscure iterator
abstraction that dates back many many years.
Move EraseDeadInstructions to Transforms/Utils and name it
RecursivelyDeleteTriviallyDeadInstructions.
llvm-svn: 60191
performance in most cases on the Grawp tester, but does speed some
things up (like shootout/hash by 15%). This also doesn't impact
compile time in a noticable way on the Grawp tester.
It also, of course, gets the testcase it was designed for right :)
llvm-svn: 60120
heuristic: the value is already live at the new memory operation if
it is used by some other instruction in the memop's block. This is
cheap and simple to compute (moreso than full liveness).
This improves the new heuristic even more. For example, it cuts two
out of three new instructions out of 255.vortex:DbmFileInGrpHdr,
which is one of the functions that the heuristic regressed. This
overall eliminates another 40 instructions from 403.gcc and visibly
reduces register pressure in 255.vortex (though this only actually
ends up saving the 2 instructions from the whole program).
llvm-svn: 60084
phrased in terms of liveness instead of as a horrible hack. :)
In pratice, this doesn't change the generated code for either
255.vortex or 403.gcc, but it could cause minor code changes in
theory. This is framework for coming changes.
llvm-svn: 60082
-enable-smarter-addr-folding to llc) that gives CGP a better
cost model for when to sink computations into addressing modes.
The basic observation is that sinking increases register
pressure when part of the addr computation has to be available
for other reasons, such as having a use that is a non-memory
operation. In cases where it works, it can substantially reduce
register pressure.
This code is currently an overall win on 403.gcc and 255.vortex
(the two things I've been looking at), but there are several
things I want to do before enabling it by default:
1. This isn't doing any caching of results, so it is much slower
than it could be. It currently slows down release-asserts llc
by 1.7% on 176.gcc: 27.12s -> 27.60s.
2. This doesn't think about inline asm memory operands yet.
3. The cost model botches the case when the needed value is live
across the computation for other reasons.
I'll continue poking at this, and eventually turn it on as llcbeta.
llvm-svn: 60074
optimize addressing modes. This allows us to optimize things like isel-sink2.ll
into:
movl 4(%esp), %eax
cmpb $0, 4(%eax)
jne LBB1_2 ## F
LBB1_1: ## TB
movl $4, %eax
ret
LBB1_2: ## F
movzbl 7(%eax), %eax
ret
instead of:
_test:
movl 4(%esp), %eax
cmpb $0, 4(%eax)
leal 4(%eax), %eax
jne LBB1_2 ## F
LBB1_1: ## TB
movl $4, %eax
ret
LBB1_2: ## F
movzbl 3(%eax), %eax
ret
This shrinks (e.g.) 403.gcc from 1133510 to 1128345 lines of .s.
Note that the 2008-10-16-SpillerBug.ll testcase is dubious at best, I doubt
it is really testing what it thinks it is.
llvm-svn: 60068
can recursively match things) and scales by 0 by ignoring them.
This triggers once in 403.gcc, saving 1 (!!!!) instruction in the
whole huge app.
llvm-svn: 60013
into a new AddressingModeMatcher class. This makes it easier
to reason about and reduces passing around of stuff, but has
no functionality change.
llvm-svn: 60012
of apint codegen failure is the DAG combiner doing
the wrong thing because it was comparing MVT's using
< rather than comparing the number of bits. Removing
the < method makes this mistake impossible to commit.
Instead, add helper methods for comparing bits and use
them.
llvm-svn: 52098
and better control the abstraction. Rename the type
to MVT. To update out-of-tree patches, the main
thing to do is to rename MVT::ValueType to MVT, and
rewrite expressions like MVT::getSizeInBits(VT) in
the form VT.getSizeInBits(). Use VT.getSimpleVT()
to extract a MVT::SimpleValueType for use in switch
statements (you will get an assert failure if VT is
an extended value type - these shouldn't exist after
type legalization).
This results in a small speedup of codegen and no
new testsuite failures (x86-64 linux).
llvm-svn: 52044
When choosing between constraints with multiple options,
like "ir", test to see if we can use the 'i' constraint and
go with that if possible. This produces more optimal ASM in
all cases (sparing a register and an instruction to load it),
and fixes inline asm like this:
void test () {
asm volatile (" %c0 %1 " : : "imr" (42), "imr"(14));
}
Previously we would dump "42" into a memory location (which
is ok for the 'm' constraint) which would cause a problem
because the 'c' modifier is not valid on memory operands.
Isn't it great how inline asm turns 'missed optimization'
into 'compile failed'??
Incidentally, this was the todo in
PowerPC/2007-04-24-InlineAsm-I-Modifier.ll
Please do NOT pull this into Tak.
llvm-svn: 50315
to the block that defines their operands. This doesn't work in the
case that the operand is an invoke, because invoke is a terminator
and must be the last instruction in a block.
Replace it with support in SelectionDAGISel for copying struct values
into sequences of virtual registers.
llvm-svn: 50279
it is only a partial fix. This change is noise for most programs, but
speeds up Shootout-C++/matrix by 20%, Ptrdist/ks by 24%, smg2000 by 8%,
hexxagon by 9%, bzip2 by 9% (not sure I trust this), ackerman by 13%, etc.
OTOH, it slows down Shootout/fib2 by 40% (I'll update PR1877 with this info).
llvm-svn: 45354
The meaning of getTypeSize was not clear - clarifying it is important
now that we have x86 long double and arbitrary precision integers.
The issue with long double is that it requires 80 bits, and this is
not a multiple of its alignment. This gives a primitive type for
which getTypeSize differed from getABITypeSize. For arbitrary precision
integers it is even worse: there is the minimum number of bits needed to
hold the type (eg: 36 for an i36), the maximum number of bits that will
be overwriten when storing the type (40 bits for i36) and the ABI size
(i.e. the storage size rounded up to a multiple of the alignment; 64 bits
for i36).
This patch removes getTypeSize (not really - it is still there but
deprecated to allow for a gradual transition). Instead there is:
(1) getTypeSizeInBits - a number of bits that suffices to hold all
values of the type. For a primitive type, this is the minimum number
of bits. For an i36 this is 36 bits. For x86 long double it is 80.
This corresponds to gcc's TYPE_PRECISION.
(2) getTypeStoreSizeInBits - the maximum number of bits that is
written when storing the type (or read when reading it). For an
i36 this is 40 bits, for an x86 long double it is 80 bits. This
is the size alias analysis is interested in (getTypeStoreSize
returns the number of bytes). There doesn't seem to be anything
corresponding to this in gcc.
(3) getABITypeSizeInBits - this is getTypeStoreSizeInBits rounded
up to a multiple of the alignment. For an i36 this is 64, for an
x86 long double this is 96 or 128 depending on the OS. This is the
spacing between consecutive elements when you form an array out of
this type (getABITypeSize returns the number of bytes). This is
TYPE_SIZE in gcc.
Since successive elements in a SequentialType (arrays, pointers
and vectors) need to be aligned, the spacing between them will be
given by getABITypeSize. This means that the size of an array
is the length times the getABITypeSize. It also means that GEP
computations need to use getABITypeSize when computing offsets.
Furthermore, if an alloca allocates several elements at once then
these too need to be aligned, so the size of the alloca has to be
the number of elements multiplied by getABITypeSize. Logically
speaking this doesn't have to be the case when allocating just
one element, but it is simpler to also use getABITypeSize in this
case. So alloca's and mallocs should use getABITypeSize. Finally,
since gcc's only notion of size is that given by getABITypeSize, if
you want to output assembler etc the same as gcc then getABITypeSize
is the size you want.
Since a store will overwrite no more than getTypeStoreSize bytes,
and a read will read no more than that many bytes, this is the
notion of size appropriate for alias analysis calculations.
In this patch I have corrected all type size uses except some of
those in ScalarReplAggregates, lib/Codegen, lib/Target (the hard
cases). I will get around to auditing these too at some point,
but I could do with some help.
Finally, I made one change which I think wise but others might
consider pointless and suboptimal: in an unpacked struct the
amount of space allocated for a field is now given by the ABI
size rather than getTypeStoreSize. I did this because every
other place that reserves memory for a type (eg: alloca) now
uses getABITypeSize, and I didn't want to make an exception
for unpacked structs, i.e. I did it to make things more uniform.
This only effects structs containing long doubles and arbitrary
precision integers. If someone wants to pack these types more
tightly they can always use a packed struct.
llvm-svn: 43620
this fixes problems where codegenprepare would sink expressions into load/stores
that are not valid, and fixes cases where it would miss important valid ones.
This fixes several serious codesize and perf issues, particularly on targets
with complex addressing modes like arm and x86. For example, now we compile
CodeGen/X86/isel-sink.ll to:
_test:
movl 8(%esp), %eax
movl 4(%esp), %ecx
cmpl $1233, %eax
ja LBB1_2 #F
LBB1_1: #T
movl $4, (%ecx,%eax,4)
movl $141, %eax
ret
LBB1_2: #F
movl (%ecx,%eax,4), %eax
ret
instead of:
_test:
movl 8(%esp), %eax
leal (,%eax,4), %ecx
addl 4(%esp), %ecx
cmpl $1233, %eax
ja LBB1_2 #F
LBB1_1: #T
movl $4, (%ecx)
movl $141, %eax
ret
LBB1_2: #F
movl (%ecx), %eax
ret
llvm-svn: 35970
isel has its own particular features that it wants in the CFG, in order to
reduce the number of times a constant is computed, etc. Make sure that we
clean up the CFG before doing any other things for isel. Doing so can
dramatically reduce the number of split edges and reduce the number of
places that constants get computed. For example, this shrinks
CodeGen/Generic/phi-immediate-factoring.ll from 44 to 37 instructions on X86,
and from 21 to 17 MBB's in the output. This is primarily a code size win,
not a performance win.
This implements CodeGen/Generic/phi-immediate-factoring.ll and PR1296.
llvm-svn: 35575
Chris Lattner