Summary:
Vectors built with zeros and elements in the same order as another
(source) vector are optimized to be built using a single insertps
instruction.
Also optimize when we move one element in a vector to a different place
in that vector while zeroing out some of the other elements.
Further optimizations are possible, described in TODO comments.
I will be implementing at least some of them in the near future.
Added some tests for different cases where this optimization triggers.
Reviewers: nadav, delena, craig.topper
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D3521
llvm-svn: 208271
The loop stream detector (LSD) on modern Intel cores, which optimizes the
execution of small loops, has limits on the number of taken branches in
addition to uop-count limits (modern AMD cores have similar limits).
Unfortunately, at the IR level, estimating the number of branches that will be
taken is difficult. For one thing, it strongly depends on later passes (block
placement, etc.). The original implementation took a conservative approach and
limited the maximal BB DFS depth of the loop. However, fairly-extensive
benchmarking by several of us has revealed that this is the wrong approach. In
fact, there are zero known cases where the branch limit prevents a detrimental
unrolling (but plenty of cases where it does prevent beneficial unrolling).
While we could improve the current branch counting logic by incorporating
branch probabilities, this further complication seems unjustified without a
motivating regression. Instead, unless and until a regression appears, the
branch counting will be removed.
llvm-svn: 208255
Given a FMA family (e.g., 213, 231), not all the variants (i.e., register or
memory) are commutable.
E.g., for the 213 family (with the syntax src1, src2, src3):
fmaXXX213 A, B, reg3/mem3 == fmaXXX213 B, A, reg3/mem3
Now consider the 231 family:
fmaXXX231 A, B, reg3 == fmaXXX231 A, reg3, B
But
fmaXXX231 A, B, mem3 != fmaXXX231 A, mem3, B
Indeed, mem3 cannot be the second argument of the memory variant of fmaXXX231.
Working on a reduced test case!
<rdar://problem/16800495>
llvm-svn: 208252
default architecture for reasonable modern x86 processors, actually be
modern. This processor model should essentially be "tuned" for modern
x86 chips as much as possible without undue penalties on any specific
architecture. Previously we weren't even using the nice scheduling
models. There are a few other tweaks needed here, but this change at
least I have benchmarked across a decent swatch of chips (intel's
clovertown, westmere, and sandybridge; amd's istanbul) and seen no
significant regressions.
If anyone has suggested ways to test this, just let me know. Somewhat
alarmingly, no existing tests failed.
llvm-svn: 208230
this patch disables the dead register elimination pass and the load/store pair
optimization pass at -O0. The ILP optimizations don't require the optimization
level to be checked because the call to addILPOpts is predicated with the
necessary check. The AdvSIMDScalar pass is disabled by default at all
optimization levels. This patch leaves that pass disabled by default.
Also, move command-line options into ARM64TargetMachine.cpp and add a few
additional flags to aid in debugging. This fixes an issue with the
-debug-pass=Structure flag where passes were printed, but not actually run
(i.e., AdvSIMDScalar pass).
llvm-svn: 208223
Summary:
These processors will only be available for the integrated assembler at
first (CodeGen will emit a fatal error saying they are not implemented).
The intention is to work through the existing instructions and correctly
annotate the ISA they were added in so that we have a sufficiently good
base to start MIPS64r6 development. MIPS64r6 removes/re-encodes certain
instructions and I believe it is best to define ISA's using set-union's
as far as possible rather than using set-subtraction.
Reviewers: vmedic
Subscribers: emaste, llvm-commits
Differential Revision: http://reviews.llvm.org/D3569
llvm-svn: 208221
When performing a scalar comparison that feeds into a vector select,
it's actually better to do the comparison on the vector side: the
scalar route would be "CMP -> CSEL -> DUP", the vector is "CM -> DUP"
since the vector comparisons are all mask based.
llvm-svn: 208210
Summary:
One small functional change. The recently added PAUSE instruction now has
the HasStdEnc predicate which was accidentally removed by a Requires<>.
Depends on D3640
Reviewers: vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3641
llvm-svn: 208209
Summary:
Move IsGP64bit into GPRPredicates, and IsFP64bit/NotFP64bit into FGRPredicates
No functional change (confirmed by diffing tablegen-erated files).
Depends on D3639
Reviewers: vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3640
llvm-svn: 208201
The AAPCS states that values passed in registers must have a value as though
they had been loaded with "LDR". LDR is equivalent to "LD1.64 vX.1D" - that is,
loading scalars to vector registers and loading 1-element vectors is equivalent.
The logic implemented here is to ensure that at all call boundaries and during
formal argument lowering all vectors are treated as their bitwidth-based floating
point scalar counterpart, which is always one of f64 or f128 (v2i32 -> f64,
v4i32 -> f128 etc). A BITCAST is inserted so that the appropriate REV will be
generated during code generation.
llvm-svn: 208198
Summary:
This makes it easier to prove a more complicated change in the next commit
is non-functional.
Reviewers: vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3639
llvm-svn: 208197
Because we've canonicalised on using LD1/ST1, every time we do a bitcast
between vector types we must do an equivalent lane reversal.
Consider a simple memory load followed by a bitconvert then a store.
v0 = load v2i32
v1 = BITCAST v2i32 v0 to v4i16
store v4i16 v2
In big endian mode every memory access has an implicit byte swap. LDR and
STR do a 64-bit byte swap, whereas LD1/ST1 do a byte swap per lane - that
is, they treat the vector as a sequence of elements to be byte-swapped.
The two pairs of instructions are fundamentally incompatible. We've decided
to use LD1/ST1 only to simplify compiler implementation.
LD1/ST1 perform the equivalent of a sequence of LDR/STR + REV. This makes
the original code sequence: v0 = load v2i32
v1 = REV v2i32 (implicit)
v2 = BITCAST v2i32 v1 to v4i16
v3 = REV v4i16 v2 (implicit)
store v4i16 v3
But this is now broken - the value stored is different to the value loaded
due to lane reordering. To fix this, on every BITCAST we must perform two
other REVs:
v0 = load v2i32
v1 = REV v2i32 (implicit)
v2 = REV v2i32
v3 = BITCAST v2i32 v2 to v4i16
v4 = REV v4i16
v5 = REV v4i16 v4 (implicit)
store v4i16 v5
This means an extra two instructions, but actually in most cases the two REV
instructions can be combined into one. For example:
(REV64_2s (REV64_4h X)) === (REV32_4h X)
There is also no 128-bit REV instruction. This must be synthesized with an
EXT instruction.
Most bitconverts require some sort of conversion. The only exceptions are:
a) Identity conversions - vNfX <-> vNiX
b) Single-lane-to-scalar - v1fX <-> fX or v1iX <-> iX
Even though there are hundreds of changed lines, I have a fairly high confidence
that they are somewhat correct. The changes to add two REV instructions per
bitcast were pretty mechanical, and once I'd done that I threw the resulting
.td at a script I wrote which combined the two REVs together (and added
an EXT instruction, for f128) based on an instruction description I gave it.
This was much less prone to error than doing it all manually, plus my brain
would not just have melted but would have vapourised.
llvm-svn: 208194
This completes the port of r204814 (cpirker "AArch64_BE function argument
passing for ARM ABI") from AArch64 to ARM64, and fixes a bunch of issues
found during later development along the way. The biggest of these was
that the alignment fixup logic wasn't replicated into all the places it
should have been.
llvm-svn: 208192
Summary:
The overall idea is to chop the Predicates list into subsets that are
usually overridden independently. This allows subclasses to partially
override the predicates of their superclasses without having to re-add all
the existing predicates.
This patch starts the process by moving HasStdEnc into a new
EncodingPredicates list and almost everything else into
AdditionalPredicates.
It has revealed a couple likely bugs where 'let Predicates' has removed
the HasStdEnc predicate.
No functional change (confirmed by diffing tablegen-erated files).
Depends on D3549, D3506
Reviewers: vmedic
Differential Revision: http://reviews.llvm.org/D3550
llvm-svn: 208184
Summary:
This will make it easier to prove that a more complicated change in the
following commit is non-functional.
No functional change.
Depends on D3506
Reviewers: vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3549
llvm-svn: 208179
Mark up additional instructions which are part of the function prologue as
MachineFrameSetup. These instructions are part of the function prologue,
emitted by the PEI pass to setup the stack for use in the activating frame.
llvm-svn: 208153
The ARM::BLX instruction is an ARM mode instruction. The Windows on ARM target
is limited to Thumb instructions. Correctly use the thumb mode tBLXr
instruction. This would manifest as an errant write into the object file as the
instruction is 4-bytes in length rather than 2. The result would be a corrupted
object file that would eventually result in an executable that would crash at
runtime.
llvm-svn: 208152
remove it from the list of unspilled registers. Otherwise the following
attempt to keep the stack aligned by picking an extra GPR register to
spill will not work as it picks up r11.
llvm-svn: 208129
Before this patch, the backend always emitted a store+load sequence to
bitconvert from f64 to i64 the input operand of a ISD::BITCAST dag node that
performed a bitconvert from type MVT::f64 to type MVT::v2i32. The resulting
i64 node was then used to build a v2i32 vector.
With this patch, the backend now produces a cheaper SCALAR_TO_VECTOR from
MVT::f64 to MVT::v2f64. That SCALAR_TO_VECTOR is then followed by a "free"
bitcast to type MVT::v4i32. The elements of the resulting
v4i32 are then extracted to build a v2i32 vector (which is illegal and
therefore promoted to MVT::v2i64).
This is in general cheaper than emitting a stack store+load sequence
to bitconvert the operand from type f64 to type i64.
llvm-svn: 208107
This patch implements the infrastructure to use named register constructs in
programs that need access to specific registers (bare metal, kernels, etc).
So far, only the stack pointer is supported as a technology preview, but as it
is, the intrinsic can already support all non-allocatable registers from any
architecture.
llvm-svn: 208104
The Win64 docs are very clear that anything larger than 8 bytes is
passed by reference, and GCC MinGW64 honors that for __modti3 and
friends.
Patch by Jameson Nash!
llvm-svn: 208029
Summary:
Also ran clang-format on the function. The code added is the last else
if block.
Reviewers: nadav, craig.topper
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D3518
llvm-svn: 207992
Windows on ARM does not conform to AEABI. However, memset would be emitted
using the AEABI signature, resulting in inverted parameters. Handle this
special case appropriately.
llvm-svn: 207943
Add handling for FK_SecRel_4 (4-byte section relative relocations). These are
used by the generation of DWARF debug information (the abbrevations use section
relative relocations). This will also be used in generation of CodeView line
tables.
llvm-svn: 207941
Filipe Cabecinhas