Previously, getConstantOffset could return an APInt with a different
bitwidth than the input pointers. For example, we might be loading an
opaque 64-bit pointer, but stripAndAccumulateInBoundsConstantOffsets
might give a 32-bit offset.
This was OK in most cases because in gatherChains, we casted the APInt
back to the original ASPtrBits.
But it was not OK when considering selects. We'd call getConstantOffset
twice and compare the resulting APInt's, which might not have the same
bit width.
This fixes that. Now getConstantOffset always returns offsets with the
correct width, so we don't need the hack of casting it in gatherChains,
and it works correctly when we're handling selects.
Differential Revision: https://reviews.llvm.org/D151640
Previously we used the later of GEPA or GEPB. This is hacky because
really we should be using the later of the two load/store instructions
being considered. But also it's flat-out incorrect, because GEPA and
GEPB might be in different BBs, in which case we cannot ask which one
comes last (assertion failure,
https://reviews.llvm.org/D149893#4378332).
Fixed, now we use the correct context instruction.
Differential Revision: https://reviews.llvm.org/D151630
The motivation for this change is a workload generated by the XLA compiler
targeting nvidia GPUs.
This kernel has a few hundred i8 loads and stores. Merging is critical for
performance.
The current LSV doesn't merge these well because it only considers instructions
within a block of 64 loads+stores. This limit is necessary to contain the
O(n^2) behavior of the pass. I'm hesitant to increase the limit, because this
pass is already one of the slowest parts of compiling an XLA program.
So we rewrite basically the whole thing to use a new algorithm. Before, we
compared every load/store to every other to see if they're consecutive. The
insight (from tra@) is that this is redundant. If we know the offset from PtrA
to PtrB, then we don't need to compare PtrC to both of them in order to tell
whether C may be adjacent to A or B.
So that's what we do. When scanning a basic block, we maintain a list of
chains, where we know the offset from every element in the chain to the first
element in the chain. Each instruction gets compared only to the leaders of
all the chains.
In the worst case, this is still O(n^2), because all chains might be of length
1. To prevent compile time blowup, we only consider the 64 most recently used
chains. Thus we do no more comparisons than before, but we have the potential
to make much longer chains.
This rewrite affects many tests. The changes to tests fall into two
categories.
1. The old code had what appears to be a bug when deciding whether a misaligned
vectorized load is fast. Suppose TTI reports that load <i32 x 4> align 4
has relative speed 1, and suppose that load i32 align 4 has relative speed
32.
The intent of the code seems to be that we prefer the scalar load, because
it's faster. But the old code would choose the vectorized load.
accessIsMisaligned would set RelativeSpeed to 0 for the scalar load (and not
even call into TTI to get the relative speed), because the scalar load is
aligned.
After this patch, we will prefer the scalar load if it's faster.
2. This patch changes the logic for how we vectorize. Usually this results in
vectorizing more.
Explanation of changes to tests:
- AMDGPU/adjust-alloca-alignment.ll: #1
- AMDGPU/flat_atomic.ll: #2, we vectorize more.
- AMDGPU/int_sideeffect.ll: #2, there are two possible locations for the call to @foo, and the pass is brittle to this. Before, we'd vectorize in case 1 and not case 2. Now we vectorize in case 2 and not case 1. So we just move the call.
- AMDGPU/adjust-alloca-alignment.ll: #2, we vectorize more
- AMDGPU/insertion-point.ll: #2 we vectorize more
- AMDGPU/merge-stores-private.ll: #1 (undoes changes from git rev 86f9117d476, which appear to have hit the bug from #1)
- AMDGPU/multiple_tails.ll: #1
- AMDGPU/vect-ptr-ptr-size-mismatch.ll: Fix alignment (I think related to #1 above).
- AMDGPU CodeGen: I have difficulty commenting on these changes, but many of them look like #2, we vectorize more.
- NVPTX/4x2xhalf.ll: Fix alignment (I think related to #1 above).
- NVPTX/vectorize_i8.ll: We don't generate <3 x i8> vectors on NVPTX because they're not legal (and eventually get split)
- X86/correct-order.ll: #2, we vectorize more, probably because of changes to the chain-splitting logic.
- X86/subchain-interleaved.ll: #2, we vectorize more
- X86/vector-scalar.ll: #2, we can now vectorize scalar float + <1 x float>
- X86/vectorize-i8-nested-add-inseltpoison.ll: Deleted the nuw test because it was nonsensical. It was doing `add nuw %v0, -1`, but this is equivalent to `add nuw %v0, 0xffff'ffff`, which is equivalent to asserting that %v0 == 0.
- X86/vectorize-i8-nested-add.ll: Same as nested-add-inseltpoison.ll
Differential Revision: https://reviews.llvm.org/D149893
This is a follow-up to b71edfaa4ec3c998aadb35255ce2f60bba2940b0
since I forgot the lit.local.cfg files in that one.
Reformatting is done with `black`.
If you end up having problems merging this commit because you
have made changes to a python file, the best way to handle that
is to run git checkout --ours <yourfile> and then reformat it
with black.
If you run into any problems, post to discourse about it and
we will try to help.
RFC Thread below:
https://discourse.llvm.org/t/rfc-document-and-standardize-python-code-style
Reviewed By: barannikov88, kwk
Differential Revision: https://reviews.llvm.org/D150762
Re-land D145441 with data layout upgrade code fixed to not break OpenMP.
This reverts commit 3f2fbe92d0f40bcb46db7636db9ec3f7e7899b27.
Differential Revision: https://reviews.llvm.org/D149776
Per discussion at
https://discourse.llvm.org/t/representing-buffer-descriptors-in-the-amdgpu-target-call-for-suggestions/68798,
we define two new address spaces for AMDGCN targets.
The first is address space 7, a non-integral address space (which was
already in the data layout) that has 160-bit pointers (which are
256-bit aligned) and uses a 32-bit offset. These pointers combine a
128-bit buffer descriptor and a 32-bit offset, and will be usable with
normal LLVM operations (load, store, GEP). However, they will be
rewritten out of existence before code generation.
The second of these is address space 8, the address space for "buffer
resources". These will be used to represent the resource arguments to
buffer instructions, and new buffer intrinsics will be defined that
take them instead of <4 x i32> as resource arguments. ptr
addrspace(8). These pointers are 128-bits long (with the same
alignment). They must not be used as the arguments to getelementptr or
otherwise used in address computations, since they can have
arbitrarily complex inherent addressing semantics that can't be
represented in LLVM. Even though, like their address space 7 cousins,
these pointers have deterministic ptrtoint/inttoptr semantics, they
are defined to be non-integral in order to prevent optimizations that
rely on pointers being a [0, [addr_max]] value from applying to them.
Future work includes:
- Defining new buffer intrinsics that take ptr addrspace(8) resources.
- A late rewrite to turn address space 7 operations into buffer
intrinsics and offset computations.
This commit also updates the "fallback address space" for buffer
intrinsics to the buffer resource, and updates the alias analysis
table.
Depends on D143437
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D145441
Many uses of getIntPtrType() were using that type to calculate the
neened type for GEP offset arguments. However, some time ago,
DataLayout was extended to support pointers where the size of the
pointer is not equal to the size of the values used to index it.
Much code was already migrated to, for example, use getIndexSizeInBits
instead of getPtrSizeInBits, but some rewrites still used
getIntPtrType() to get the type for GEP offsets.
This commit changes uses of getIntPtrType() to getIndexType() where
they are involved in a GEP-related calculation.
In at least one case (bounds check insertion) this resolves a compiler
crash that the new test added here would previously trigger.
This commit does not impact
- C library-related rewriting (memcpy()), which are operating under
the assumption that intptr_t == size_t. While all the mechanisms for
breaking this assumption now exist, doing so is outside the scope of
this commit.
- Code generation and below. Note that the use of getIntPtrType() in
CodeGenPrepare will be changed in a future commit.
- Usage of getIntPtrType() in any backend
Depends on D143435
Reviewed By: arichardson
Differential Revision: https://reviews.llvm.org/D143437
Based on experimentation on gfx906,908,90a and 1030, wider global loads / stores are more performant than multiple narrower ones independent of alignment -- this is especially true when combining 8 bit loads / stores, in which case speedup was usually 2x across all alignments.
Differential Revision: https://reviews.llvm.org/D145170
Change-Id: I6ee6c76e6ace7fc373cc1b2aac3818fc1425a0c1
This patch is changing the InsertElement's placeholder to poison without changing the LSV's behavior.
Regardless of whether `StoreTy` is FixedVectorType or not, the poison value will be overwritten with a different value.
Therefore, whether the InsertElement's placeholder is poison or undef will not affect the result of the program.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D111005
This commit copies existing tests at llvm/Transforms and replaces
'insertelement undef' in those files with 'insertelement poison'.
(see https://reviews.llvm.org/D93586)
Tests listed using this script:
grep -R -E '^[^;]*insertelement <.*> undef,' . | cut -d":" -f1 | uniq |
wc -l
Tests updated:
file_org=llvm/test/Transforms/$1
file=${file_org%.ll}-inseltpoison.ll
cp $file_org $file
sed -i -E 's/^([^;]*)insertelement <(.*)> undef/\1insertelement <\2> poison/g' $file
head -1 $file | grep "Assertions have been autogenerated by utils/update_test_checks.py" -q
if [ "$?" == 1 ]; then
echo "$file : should be manually updated"
# I manually updated the script
exit 1
fi
python3 ./llvm/utils/update_test_checks.py --opt-binary=./build-releaseassert/bin/opt $file
Adjust SITargetLowering::allowsMisalignedMemoryAccessesImpl for
unaligned flat scratch support. Mostly needed for global isel.
Differential Revision: https://reviews.llvm.org/D93669
Explicitly opt-out llvm/test/Transforms/Attributor.
Verified by flipping the default value of allow-unused-prefixes and
observing that none of the failures were under llvm/test/Transforms.
Differential Revision: https://reviews.llvm.org/D92404
Features UnalignedBufferAccess and UnalignedDSAccess are now used to determine
whether hardware supports such access.
UnalignedAccessMode should be used to enable them.
hasUnalignedBufferAccessEnabled() and hasUnalignedDSAccessEnabled() can be
now used to quickly check both.
Differential Revision: https://reviews.llvm.org/D84522
Adjust alignment requirements for ds_read/write_b96/b128.
GFX9 and onwards allow misaligned access for reads and writes but only if
SH_MEM_CONFIG.alignment_mode allows it.
UnalignedDSAccess is set on GCN subtargets from GFX9 onward to let us know if we
can relax alignment requirements.
UnalignedAccessMode acts similary to UnalignedBufferAccess for DS instructions
but only from GFX9 onward and is supposed to match alignment_mode. By default
alignment of 4 is required.
Differential Revision: https://reviews.llvm.org/D82788
Summary: To match NewPM name. Also the new name is clearer and more consistent.
Subscribers: jvesely, nhaehnle, hiraditya, asbirlea, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84542
This is apparently worse than 1-byte alignment. This does not attempt
to decompose 2-byte aligned wide stores, but will stop trying to
produce them.
Also fix bug in LoadStoreVectorizer which was decreasing the alignment
and vectorizing stack accesses. It was assuming a stack object was an
alloca that could have its base alignment changed, which is not true
if the pointer is derived from a function argument.
Added code to truncate or shrink offsets so that we can continue
base pointer search if size has changed along the way.
Differential Revision: https://reviews.llvm.org/D65612
llvm-svn: 367646
The previous change to fix crash in the vectorizer introduced
performance regressions. The condition to preserve pointer
address space during the search is too tight, we only need to
match the size.
Differential Revision: https://reviews.llvm.org/D65600
llvm-svn: 367624
When vectorizer strips pointers it can eventually end up with
pointers of two different sizes, then SCEV will crash.
Differential Revision: https://reviews.llvm.org/D65480
llvm-svn: 367443
As it's causing some bot failures (and per request from kbarton).
This reverts commit r358543/ab70da07286e618016e78247e4a24fcb84077fda.
llvm-svn: 358546
If the alignment is at least 4, this should report true.
Something still seems off with how < 4-byte types are
handled here though.
Fixing this seems to change how some combines get
to where they get, but somehow isn't changing the net
result.
llvm-svn: 342879
This was checking the hardcoded address space 0 for the stack.
Additionally, this should be checking for legality with
the adjusted alignment, so defer the alignment check.
Also try to split if the unaligned access isn't allowed.
llvm-svn: 342442
In some cases LSV sees (load/store _ (select _ <pointer expression>
<pointer expression>)) patterns in input IR, often due to sinking and
other forms of CFG simplification, sometimes interspersed with
bitcasts and all-constant-indices GEPs. With this
patch`areConsecutivePointers` method would attempt to handle select
instructions. This leads to an increased number of successful
vectorizations.
Technically, select instructions could appear in index arithmetic as
well, however, we don't see those in our test suites / benchmarks.
Also, there is a lot more freedom in IR shapes computing integral
indices in general than in what's common in pointer computations, and
it appears that it's quite unreliable to do anything short of making
select instructions first class citizens of Scalar Evolution, which
for the purposes of this patch is most definitely an overkill.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D49428
llvm-svn: 337965
This reapplies commit r337489 reverted by r337541
Additionally, this commit contains a speculative fix to the issue reported in r337541
(the report does not contain an actionable reproducer, just a stack trace)
llvm-svn: 337606
This is mostly a preparation work for adding a limited support for
select instructions. It proved to be difficult to do due to size and
irregularity of Vectorizer::isConsecutiveAccess, this is fixed here I
believe.
It also turned out that these changes make it simpler to finish one of
the TODOs and fix a number of other small issues, namely:
1. Looking through bitcasts to a type of a different size (requires
careful tracking of the original load/store size and some math
converting sizes in bytes to expected differences in indices of GEPs).
2. Reusing partial analysis of pointers done by first attempt in proving
them consecutive instead of starting from scratch. This added limited
support for nested GEPs co-existing with difficult sext/zext
instructions. This also required a careful handling of negative
differences between constant parts of offsets.
3. Handing a case where the first pointer index is not an add, but
something else (a function parameter for instance).
I observe an increased number of successful vectorizations on a large
set of shader programs. Only few shaders are affected, but those that
are affected sport >5% less loads and stores than before the patch.
Reviewed By: rampitec
Differential-Revision: https://reviews.llvm.org/D49342
llvm-svn: 337489
Summary: Currently, isConsecutiveAccess() detects two pointers(PtrA and PtrB) as consecutive by
comparing PtrB with BaseDelta+PtrA. This works when both pointers are factorized or
both of them are not factorized. But isConsecutiveAccess() fails if one of the
pointers is factorized but the other one is not.
Here is an example:
PtrA = 4 * (A + B)
PtrB = 4 + 4A + 4B
This patch uses getMinusSCEV() to compute the distance between two pointers.
getMinusSCEV() allows combining the expressions and computing the simplified distance.
Author: FarhanaAleen
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D49516
llvm-svn: 337471
The memory location an invariant load is using can never be clobbered by
any store, so it's safe to move the load ahead of the store.
Differential Revision: https://reviews.llvm.org/D46011
llvm-svn: 330725
Author: Samuel Pitoiset
ds_read_b128 and ds_write_b128 have been recently enabled
under the amdgpu-ds128 option because the performance benefit
is unclear.
Though, using 128-bit loads/stores for the local address space
appears to introduce regressions in tessellation shaders. Not
sure what is broken, but as ds_read_b128/ds_write_b128 are not
enabled by default, just introduce a global option and enable
128-bit only if requested (until it's fixed/used correctly).
v2: - fix regressions in merge-stores.ll and multiple_tails.ll
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105464
llvm-svn: 329764
Summary: Starting from GCN 2nd generation, ISA supports ds_read_b128 on top of ds_read_b64.
This patch supports ds_read_b128 instruction pattern and generation of this instruction.
In the vectorizer, this patch also widen the vector length so that vectorizer generates
128 bit loads for local address-space which gets translated to ds_read_b128.
Since the performance benefit is not clear; compiler generates ds_read_b128 under -amdgpu-ds128.
Author: FarhanaAleen
Reviewed By: rampitec, arsenm
Subscribers: llvm-commits, AMDGPU
Differential Revision: https://reviews.llvm.org/D44210
llvm-svn: 327153
