In order to facilitate targets that only support masked loads/stores
on certain address spaces (AMDGPU will support them in an upcoming
patch, but only for address space 7), add an AddressSpace parameter
to isLegalMaskedLoad and isLegalMaskedStore
In the implementation of the getExtendedReductionCost(), it ofter calls
getArithmeticReductionCost() with FMFs. But we shouldn't call
getArithmeticReductionCost() with FMFs for non-floating-point reductions
which will return the wrong cost.
This patch makes FMFs in getExtendedReductionCost() optional and align
to the getArithmeticReductionCost(). So the TTI will return the correct
cost for non-FP extended-reductions query without FMFs.
This patch is not quite NFC but it's hard to test from the CostModel
side.
Split from #113903.
This caused assertion failures:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp:16237:
Value *llvm::slpvectorizer::BoUpSLP::vectorizeTree(TreeEntry *):
Assertion `OpTE1.isSame( ArrayRef(E->Scalars).take_front(OpTE1.getVectorFactor())) && "Expected same first part of scalars."' failed.
See comment on the PR.
> Previous version was reviewed here https://github.com/llvm/llvm-project/pull/123360
> It is mostly the same, adjusted after graph-to-tree transformation
This reverts commit 7de895ff1146c17ec78877900c01c09f4140e692.
This caused failures such as:
Instruction does not dominate all uses!
%29 = insertelement <8 x i64> %28, i64 %xor6.i.5, i64 6
%17 = shufflevector <8 x i64> %29, <8 x i64> poison, <6 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6>
see comment on https://github.com/llvm/llvm-project/pull/123360
> Previous version was reviewed here https://github.com/llvm/llvm-project/pull/123360
> It is mostly the same, adjusted after graph-to-tree transformation
>
> Patch tries to remove wide alternate operations.
> Currently SLP vectorizer emits something like this:
> ```
> %0 = add i32
> %1 = sub i32
> %2 = add i32
> %3 = sub i32
> %4 = add i32
> %5 = sub i32
> %6 = add i32
> %7 = sub i32
>
> transformes to
>
> %v1 = add <8 x i32>
> %v2 = sub <8 x i32>
> %res = shuffle %v1, %v2, <0, 9, 2, 11, 4, 13, 6, 15>
> ```
> i.e. half of the results are just unused. This leads to increased
> register pressure and potentially doubles number of operations.
>
> Patch introduces SplitVectorize mode, where it splits the operations by
> opcodes and produces instead something like this:
> ```
> %v1 = add <4 x i32>
> %v2 = sub <4 x i32>
> %res = shuffle %v1, %v2, <0, 4, 1, 5, 2, 6, 3, 7>
> ```
> It allows to improve the performance by reducing number of ops. Also, it
> turns on some other improvements, like improved graph reordering.
>
> [...]
This reverts commit 9d37e61fc77d3d6de891c30630f1c0227522031d as well as
the follow-up commit 72bb0a9a9c6fdde43e1e191f2dc0d5d2d46aff4e.
No in-tree targets currently use it in the
preferInLoopReduction/preferPredicatedReductionSelect TTI hooks. It
looks like it used to be used in LoopUtils, at least in
8ca60db40bd944dc5f67e0f200a403b4e03818ea, but I presume it was replaced
by RecurrenceDescriptor.
This patch updates the cost model to cost intrinsics that return
multiple values (in structs) correctly. Previously, the cost model only
thought intrinsics that return `VectorType` need scalarizing, which
meant it cost intrinsics that return multiple vectors (that need
scalarizing) way too cheap (giving it the cost of a single function
call).
This patch also adds a custom cost for llvm.sincos when a vector
function library is available, as certain VFs can be expanded (later in
code gen) to a vector function, reducing the cost to a single call (+
the possible loads from the vector function returns values via output
pointers).
Pulled out of #122236, this allows Splats constants to be recognized by
getOperandInfo, allowing "better" costs for instructions like divides by
constants to be produced (which are expanded into mul+add+shift). Some
of the costs are not very accurate yet, but the comparison of scalar vs
fixed-width vs scalable for the same div can become more accurate,
especially with patches like #122236.
This patch adds initial support for vectorizing literal struct return
values. Currently, this is limited to the case where the struct is
homogeneous (all elements have the same type) and not packed. The users
of the call also must all be `extractvalue` instructions.
The intended use case for this is vectorizing intrinsics such as:
```
declare { float, float } @llvm.sincos.f32(float %x)
```
Mapping them to structure-returning library calls such as:
```
declare { <4 x float>, <4 x float> } @Sleef_sincosf4_u10advsimd(<4 x float>)
```
Or their widened form (such as `@llvm.sincos.v4f32` in this case).
Implementing this required two main changes:
1. Supporting widening `extractvalue`
2. Adding support for vectorized struct types in LV
* This is mostly limited to parts of the cost model and scalarization
Since the supported use case is narrow, the required changes are
relatively small.
This patch adds methods for cost estimation for
llvm.masked.expandload/llvm.masked.compressstore intrinsics in TTI. If
backend doesn't support custom lowering of these intrinsics it will be
processed by ScalarizeMaskedMemIntrin so we estimate its cost via
getCommonMaskedMemoryOpCost as gather/scatter operation; for RISC-V
backend, this patch implements custom hook to calculate the cost based
on current lowering scheme.
Patch tries to remove wide alternate operations.
Currently SLP vectorizer emits something like this:
```
%0 = add i32
%1 = sub i32
%2 = add i32
%3 = sub i32
%4 = add i32
%5 = sub i32
%6 = add i32
%7 = sub i32
transformes to
%v1 = add <8 x i32>
%v2 = sub <8 x i32>
%res = shuffle %v1, %v2, <0, 9, 2, 11, 4, 13, 6, 15>
```
i.e. half of the results are just unused. This leads to increased
register pressure and potentially doubles number of operations.
Patch introduces SplitVectorize mode, where it splits the operations by
opcodes and produces instead something like this:
```
%v1 = add <4 x i32>
%v2 = sub <4 x i32>
%res = shuffle %v1, %v2, <0, 4, 1, 5, 2, 6, 3, 7>
```
It allows to improve the performance by reducing number of ops. Also, it
turns on some other improvements, like improved graph reordering.
-O3+LTO, AVX512
Metric: size..text
Program size..text
results results0 diff
test-suite :: MultiSource/Benchmarks/Prolangs-C/TimberWolfMC/timberwolfmc.test 277800.00 280536.00 1.0%
test-suite :: MultiSource/Benchmarks/FreeBench/pifft/pifft.test 81802.00 82426.00 0.8%
test-suite :: External/SPEC/CINT2006/464.h264ref/464.h264ref.test 790552.00 790952.00 0.1%
test-suite :: MultiSource/Applications/JM/ldecod/ldecod.test 383795.00 383987.00 0.1%
test-suite :: External/SPEC/CINT2017speed/600.perlbench_s/600.perlbench_s.test 2075541.00 2076501.00 0.0%
test-suite :: External/SPEC/CINT2017rate/500.perlbench_r/500.perlbench_r.test 2075541.00 2076501.00 0.0%
test-suite :: MultiSource/Benchmarks/Bullet/bullet.test 312702.00 312766.00 0.0%
test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test 12569783.00 12569751.00 -0.0%
test-suite :: External/SPEC/CFP2017rate/510.parest_r/510.parest_r.test 2049374.00 2049358.00 -0.0%
test-suite :: External/SPEC/CINT2006/400.perlbench/400.perlbench.test 1091836.00 1091772.00 -0.0%
test-suite :: MultiSource/Applications/JM/lencod/lencod.test 852339.00 852211.00 -0.0%
test-suite :: MultiSource/Applications/oggenc/oggenc.test 190651.00 190523.00 -0.1%
test-suite :: MultiSource/Benchmarks/DOE-ProxyApps-C/miniGMG/miniGMG.test 44203.00 44155.00 -0.1%
test-suite :: SingleSource/UnitTests/Vector/AVX512BWVL/Vector-AVX512BWVL-mask_set_bw.test 12997.00 12981.00 -0.1%
test-suite :: External/SPEC/CINT2017speed/625.x264_s/625.x264_s.test 668971.00 658427.00 -1.6%
test-suite :: External/SPEC/CINT2017rate/525.x264_r/525.x264_r.test 668971.00 658427.00 -1.6%
Prolangs-C/TimberWolfMC/timberwolfmc - small variations, some code not
inlined
FreeBench/pifft - extra stores <8 x double> vectorized, some other extra
vectorizations
CINT2006/464.h264ref - some smaller code + changes similar to x264
JM/ldecod - changes similar x264
CINT2017speed/600.perlbench_s
CINT2017rate/500.perlbench_r - significantly compact vector code
Benchmarks/Bullet - small variations
CFP2017rate/526.blender_r - small variations
CFP2017rate/510.parest_r - small variations
CINT2006/400.perlbench - extra vector code
JM/lencod - extra store <16 x i32> and other changes similar x264
Applications/oggenc - extra store <16 x i8>, small variations
DOE-ProxyApps-C/miniGMG - small variations
Vector/AVX512BWVL/Vector-AVX512BWVL-mask_set_bw - better vector code
CINT2017speed/625.x264_s
CINT2017rate/525.x264_r - the number of instructions increased, but
looks like they are more performant. E.g., for function
x264_pixel_satd_8x8, llvm-mca reports better throughput - 84 for the
current version and 59 for the new version.
-O3+LTO, march=rva32u64
CINT2017rate/525.x264_r - similar to x86, extra code in pixel_hadamard_ac
function vectorized, idct4x4dc stopped being vectorized (looks like
issue with shuffles cost)
CINT2006/400.perlbench - better vector code
CINT2006/445.gobmk - some variations in vector code
CINT2006/464.h264ref - extra code vectorized
CINT2017rate/500.perlbench_r - small variations
-O3+LTO, mcpu=sifive-p470
Metric: size..text
Program size..text
results results0 diff
test-suite :: External/SPEC/CINT2006/464.h264ref/464.h264ref.test 587336.00 587668.00 0.1%
test-suite :: MultiSource/Applications/JM/lencod/lencod.test 643308.00 643614.00 0.0%
test-suite :: MultiSource/Applications/d/make_dparser.test 79678.00 79710.00 0.0%
test-suite :: MultiSource/Benchmarks/Bullet/bullet.test 277322.00 277420.00 0.0%
test-suite :: External/SPEC/CINT2006/400.perlbench/400.perlbench.test 933660.00 933682.00 0.0%
test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test 9497722.00 9497682.00 -0.0%
test-suite :: External/SPEC/CINT2017rate/500.perlbench_r/500.perlbench_r.test 1767806.00 1767772.00 -0.0%
test-suite :: External/SPEC/CINT2017speed/600.perlbench_s/600.perlbench_s.test 1767806.00 1767772.00 -0.0%
test-suite :: MultiSource/Benchmarks/MiBench/consumer-lame/consumer-lame.test 148038.00 148024.00 -0.0%
test-suite :: MultiSource/Applications/JM/ldecod/ldecod.test 283036.00 283008.00 -0.0%
test-suite :: MultiSource/Benchmarks/mediabench/g721/g721encode/encode.test 4776.00 4772.00 -0.1%
test-suite :: External/SPEC/CINT2017rate/525.x264_r/525.x264_r.test 540582.00 511772.00 -5.3%
test-suite :: External/SPEC/CINT2017speed/625.x264_s/625.x264_s.test 540582.00 511772.00 -5.3%
CINT2006/464.h264ref - extra vector code in find_sad_16x16
JM/lencod - extra vector code in find_sad_16x16
d/make_dparser - smaller vector code
Benchmarks/Bullet - small variations
CINT2006/400.perlbench - smaller vector code
CFP2017rate/526.blender_r - small variations, extra store <8 x float> in
the loop, extra store <8 x i8> in loop
CINT2017rate/500.perlbench_r
CINT2017speed/600.perlbench_s - small variations
MiBench/consumer-lame - small variations
JM/ldecod - extra vector code
mediabench/g721/g721encode - small variations
CINT2017rate/525.x264_r
CINT2017speed/625.x264_s - reduced number of wide operations and
shuffles, saving the registers, similar to X86, extra code in
pixel_hadamard_ac vectorized, idct4x4dc not vectorized (issue with some
TTI costs)
Reviewers: RKSimon, hiraditya
Reviewed By: RKSimon
Pull Request: https://github.com/llvm/llvm-project/pull/123360
This patch implements an LLVM IR pass, named kernel-info, that reports
various statistics for codes compiled for GPUs. The ultimate goal of
these statistics to help identify bad code patterns and ways to mitigate
them. The pass operates at the LLVM IR level so that it can, in theory,
support any LLVM-based compiler for programming languages supporting
GPUs. It has been tested so far with LLVM IR generated by Clang for
OpenMP offload codes targeting NVIDIA GPUs and AMD GPUs.
By default, the pass runs at the end of LTO, and options like
``-Rpass=kernel-info`` enable its remarks. Example `opt` and `clang`
command lines appear in `llvm/docs/KernelInfo.rst`. Remarks include
summary statistics (e.g., total size of static allocas) and individual
occurrences (e.g., source location of each alloca). Examples of its
output appear in tests in `llvm/test/Analysis/KernelInfo`.
To deduce whether the optimization is legal we need to compare the target
features between caller and callee versions. The criteria for bypassing
the resolver are the following:
* If the callee's feature set is a subset of the caller's feature set,
then the callee is a candidate for direct call.
* Among such candidates the one of highest priority is the best match
and it shall be picked, unless there is a version of the callee with
higher priority than the best match which cannot be picked from a
higher priority caller (directly or through the resolver).
* For every higher priority callee version than the best match, there
is a higher priority caller version whose feature set availability
is implied by the callee's feature set.
Example:
Callers and Callees are ordered in decreasing priority.
The arrows indicate successful call redirections.
Caller Callee Explanation
=========================================================================
mops+sve2 --+--> mops all the callee versions are subsets of the
| caller but mops has the highest priority
|
mops --+ sve2 between mops and default callees, mops wins
sve sve between sve and default callees, sve wins
but sve2 does not have a high priority caller
default -----> default sve (callee) implies sve (caller),
sve2(callee) implies sve (caller),
mops(callee) implies mops(caller)
This relands the reverted #120721 with a fix for cases where neither
reduction operand are the reduction phi. Only
63114239cc8d26225a0ef9920baacfc7cc00fc58 and
63114239cc8d26225a0ef9920baacfc7cc00fc58 are new on top of the reverted
PR.
---------
Co-authored-by: Nicholas Guy <nicholas.guy@arm.com>
This re-lands the reverted #92418
When the VF is small enough so that dividing the VF by the scaling
factor results in 1, the reduction phi execution thinks the VF is scalar
and sets the reduction's output as a scalar value, tripping assertions
expecting a vector value. The latest commit in this PR fixes that by
using `State.VF` in the scalar check, rather than the divided VF.
---------
Co-authored-by: Nicholas Guy <nicholas.guy@arm.com>
- update `VectorUtils:isVectorIntrinsicWithScalarOpAtArg` to use TTI for
all uses, to allow specifiction of target specific intrinsics
- add TTI to the `isVectorIntrinsicWithStructReturnOverloadAtField` api
- update TTI api to provide `isTargetIntrinsicWith...` functions and
consistently name them
- move `isTriviallyScalarizable` to VectorUtils
- update all uses of the api and provide the TTI parameter
Resolves#117030
Following on from https://github.com/llvm/llvm-project/pull/94499, this
patch adds support to the Loop Vectorizer to emit the partial reduction
intrinsics where they may be beneficial for the target.
---------
Co-authored-by: Samuel Tebbs <samuel.tebbs@arm.com>
As vector element loads are free on SystemZ, this patch improves the cost
computation in getGatherCost() to reflect this.
getScalarizationOverhead() gets an optional parameter which can hold the actual
Values so that they in turn can be passed (by BasicTTIImpl) to
getVectorInstrCost().
SystemZTTIImpl::getVectorInstrCost() will now recognize a LoadInst and
typically return a 0 cost for it, with some exceptions.
This changes allows target intrinsics to specify and overwrite overloaded types.
- Updates `ReplaceWithVecLib` to not provide TTI as there most probably won't be a use-case
- Updates `SLPVectorizer` to use available TTI
- Updates `VPTransformState` to pass down TTI
- Updates `VPlanRecipe` to use passed-down TTI
This change will let us add scalarization for `asdouble`: #114847
Set the maximum interleaving factor to 4, aligning with the number of available
SIMD pipelines. This increases the number of vector instructions in the vectorised
loop body, enhancing performance during its execution. However, for very low
iteration counts, the vectorised body might not execute at all, leaving only the
epilogue loop to run. This issue affects e.g. cam4_r from SPEC FP, which
experienced a performance regression. To address this, the patch reduces the
minimum epilogue vectorisation factor from 16 to 8, enabling the epilogue to be
vectorised and largely mitigating the regression.
…er possible
In case of Neon, if there exists extractelement from lane != 0 such that
1. extractelement does not necessitate a move from vector_reg -> GPR
2. extractelement result feeds into fmul
3. Other operand of fmul is a scalar or extractelement from lane 0 or
lane equivalent to 0
then the extractelement can be merged with fmul in the backend and it
incurs no cost.
e.g.
```
define double @foo(<2 x double> %a) {
%1 = extractelement <2 x double> %a, i32 0
%2 = extractelement <2 x double> %a, i32 1
%res = fmul double %1, %2
ret double %res
}
```
`%2` and `%res` can be merged in the backend to generate:
`fmul d0, d0, v0.d[1]`
The change was tested with SPEC FP(C/C++) on Neoverse-v2.
**Compile time impact**: None
**Performance impact**: Observing 1.3-1.7% uplift on lbm benchmark with -flto depending upon the config.
This is a recommit of #107120 . The original PR was approved but failed
buildbot. The newly added tests should only be run for compilers that
support the ARM target. This has been resolved by adding a config file
for these tests.
- Pass optimizes memcpy's by padding out destinations and sources to a
full word to make ARM backend generate full word loads instead of
loading a single byte (ldrb) and/or half word (ldrh). Only pads
destination when it's a stack allocated constant size array and source
when it's constant string. Heuristic to decide whether to pad or not
is very basic and could be improved to allow more examples to be
padded.
- Pass works at the midend level
- Pass optimizes memcpy's by padding out destinations and sources to a
full word to make backend generate full word loads instead of loading a
single byte (ldrb) and/or half word (ldrh). Only pads destination when
it's a stack allocated constant size array and source when it's constant
array. Heuristic to decide whether to pad or not is very basic and could
be improved to allow more examples to be padded.
- Pass works within GlobalOpt but is disabled by default on all targets
except ARM.
A call to a function that has this attribute is not a source of
divergence, as used by UniformityAnalysis. That allows a front-end to
use known-name calls as an instruction extension mechanism (e.g.
https://github.com/GPUOpen-Drivers/llvm-dialects ) without such a call
being a source of divergence.
Currently we will not be able to inline a large function even if it only
has one live use because the inline cost is still very high after
applying `LastCallToStaticBonus`, which is a constant. This could
significantly impact the performance because CSR spill is very
expensive.
This PR adds a new function `getInliningLastCallToStaticBonus` to TTI to
allow targets to customize this value.
Fixes SWDEV-471398.
Porting to TTI provides direct access to the instruction cost model,
which can enable instruction cost based sinking without introducing code
duplication.
This follows in the spirit of 7d82c99403f615f6236334e698720bf979959704,
and extends the costing API for compares and selects to provide
information about the operands passed in an analogous manner. This
allows us to model the cost of materializing the vector constant, as
some select-of-constants are significantly more expensive than others
when you account for the cost of materializing the constants involved.
This is a stepping stone towards fixing
https://github.com/llvm/llvm-project/issues/109466. A separate SLP patch
will be required to utilize the new API.
Since we are using the Scalarizer pass in the backend we needed a way to
allow this pass to operate on Target intrinsics.
We achieved this by adding `TargetTransformInfo ` to the Scalarizer
pass. This allowed us to call a function available to the DirectX
backend to know if an intrinsic is a target intrinsic that should be
scalarized.
This transformation doesn't actually use any of the internal state of
LSR and recomputes all information from SCEV. Splitting it out makes
it easier to test.
Note that long term I would like to write a version of this transform
which *is* integrated with LSR's solver, but if that happens, we'll
just delete the extra pass.
Integration wise, I switched from using TTI to using a pass configuration
variable. This seems slightly more idiomatic, and means we don't run
the extra logic on any target other than RISCV.
This reverts commit e8ad87c7d06afe8f5dde2e4c7f13c314cb3a99e9.
This reverts commit d3c9bb0cf811424dcb8c848cf06773dbdde19965.
A few buildbots trip up on asan-rvv-intrinsics.ll. I've also reverted
the follow-up commit d3c9bb0cf8.
https://lab.llvm.org/buildbot/#/builders/46/builds/2895
Previously asan considers target intrinsics as black boxes, so asan
could not instrument accurate check. This patch provide TTI hooks to
make targets describe their intrinsic informations to asan.
Note,
1. this patch renames InterestingMemoryOperand to MemoryRefInfo.
2. this patch does not support RVV indexed/segment load/store.
The `!unpredictable` metadata has been present for a long time, but
it's usage in optimizations is still limited. This patch teaches
`FoldTwoEntryPHINode()` to be more aggressive with an unpredictable
branch to reduce mispredictions.
A TTI interface `getBranchMispredictPenalty()` is added to distinguish
between different hardwares to ensure we don't go too far for simpler
cores. For simplicity, only a naive x86 implementation is included for
the time being.
For the Neoverse V2 we would like to prefer fixed width over scalable
vectorisation if the cost-model assigns an equal cost to both for certain
loops. This improves 7 kernels from TSVC-2 and several production kernels by
about 2x, and does not affect SPEC21017 INT and FP. This also adds a new TTI
hook that can steer the loop vectorizater to preferring fixed width
vectorization, which can be set per CPU. For now, this is only enabled for the
Neoverse V2.
There are 3 reasons why preferring NEON might be better in the case the
cost-model is a tie and the SVE vector size is the same as NEON (128-bit):
architectural reasons, micro-architecture reasons, and SVE codegen reasons. The
latter will be improved over time, so the more important reasons are the former
two. I.e., (micro) architecture reason is the use of LPD/STP instructions which
are not available in SVE2 and it avoids predication.
For what it is worth: this codegen strategy to generate more NEON is inline
with GCC's codegen strategy, which is actually even more aggressive in
generating NEON when no predication is required. We could be smarter about the
decision making, but this seems to be a first good step in the right direction,
and we can always revise this later (for example make the target hook more
general).
WebAssembly doesn't support horizontal operations nor does it have a way
of expressing fast-math or reassoc flags, so runtimes are currently
unable to use pairwise operations when generating code from the existing
shuffle patterns.
This patch allows the backend to select which, arbitary, shuffle pattern
to be used per reduction intrinsic. The default behaviour is the same as
the existing, which is by splitting the vector into a top and bottom
half. The other pattern introduced is for a pairwise shuffle.
WebAssembly enables pairwise reductions for int/fp add/sub.
1. Add TTI interface for conditional load/store.
2. Mark 1 x i16/i32/i64 masked load/store legal so that it's not
legalized in pass scalarize-masked-mem-intrin.
3. Visit 1 x i16/i32/i64 masked load/store to build a target-specific
CLOAD/CSTORE node to avoid error in
`DAGTypeLegalizer::ScalarizeVectorResult`.
4. Combine DAG to simplify the nodes for CLOAD/CSTORE.
5. Lower CLOAD/CSTORE to CFCMOV by pattern match.
This is CodeGen part of #95515
