Since there is no instruction for integer vector division, factor in the
cost of singling out each element to be used with the scalar division
instruction.
Differential revision: https://reviews.llvm.org/D43974
llvm-svn: 326955
Agner's tables indicate that for SSE42+ targets (Core2 and later) we can reduce the FADD/FSUB/FMUL costs down to 1, which should fix the Himeno benchmark.
Note: the AVX512 FDIV costs look rather dodgy, but this isn't part of this patch.
Differential Revision: https://reviews.llvm.org/D43733
llvm-svn: 326133
There are too many perf regressions resulting from this, so we need to
investigate (and add tests for) targets like ARM and AArch64 before
trying to reinstate.
llvm-svn: 325658
This change was mentioned at least as far back as:
https://bugs.llvm.org/show_bug.cgi?id=26837#c26
...and I found a real program that is harmed by this:
Himeno running on AMD Jaguar gets 6% slower with SLP vectorization:
https://bugs.llvm.org/show_bug.cgi?id=36280
...but the change here appears to solve that bug only accidentally.
The div/rem costs for x86 look very wrong in some cases, but that's already true,
so we can fix those in follow-up patches. There's also evidence that more cost model
changes are needed to solve SLP problems as shown in D42981, but that's an independent
problem (though the solution may be adjusted after this change is made).
Differential Revision: https://reviews.llvm.org/D43079
llvm-svn: 325515
Summary:
There are few oddities that occur due to v1i1, v8i1, v16i1 being legal without v2i1 and v4i1 being legal when we don't have VLX. Particularly during legalization of v2i32/v4i32/v2i64/v4i64 masked gather/scatter/load/store. We end up promoting the mask argument to these during type legalization and then have to widen the promoted type to v8iX/v16iX and truncate it to get the element size back down to v8i1/v16i1 to use a 512-bit operation. Since need to fill the upper bits of the mask we have to fill with 0s at the promoted type.
It would be better if we could just have the v2i1/v4i1 types as legal so they don't undergo any promotion. Then we can just widen with 0s directly in a k register. There are no real v4i1/v2i1 instructions anyway. Everything is done on a larger register anyway.
This also fixes an issue that we couldn't implement a masked vextractf32x4 from zmm to xmm properly.
We now have to support widening more compares to 512-bit to get a mask result out so new tablegen patterns got added.
I had to hack the legalizer for widening the operand of a setcc a bit so it didn't try create a setcc returning v4i32, extract from it, then try to promote it using a sign extend to v2i1. Now we create the setcc with v4i1 if the original setcc's result type is v2i1. Then extract that and don't sign extend it at all.
There's definitely room for improvement with some follow up patches.
Reviewers: RKSimon, zvi, guyblank
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41560
llvm-svn: 321967
Based on the names of the check lines, features seems more appropriate that cpu.
Spotted while prototyping my patch to make 512-bit vectors illegal on SKX sometimes.
llvm-svn: 320959
This patch contains more accurate cost of interelaved load\store of stride 2 for the types int64\double on AVX2.
Reviewers: delena, RKSimon, craig.topper, dorit
Reviewed By: dorit
Differential Revision: https://reviews.llvm.org/D40008
llvm-svn: 318385
Recommit:
This patch contains update of the costs of interleaved loads of v8f32 of stride 3 and 8.
fixed the location of the lit test it works with make check-all.
Differential Revision: https://reviews.llvm.org/D39403
llvm-svn: 317471
reverted my changes will be committed later after fixing the failure
This patch contains update of the costs of interleaved loads of v8f32 of stride 3 and 8.
Differential Revision: https://reviews.llvm.org/D39403
llvm-svn: 317433
This patch contains update of the costs of interleaved loads of v8f32 of stride 3 and 8.
Differential Revision: https://reviews.llvm.org/D39403
llvm-svn: 317432
This patch adds accurate instructions cost.
The formula presents two cases(stride 3 and stride 4) and calculates the cost according to the VF and stride.
Reviewers:
1. delena
2. Farhana
3. zvi
4. dorit
5. Ayal
Differential Revision: https://reviews.llvm.org/D38762
Change-Id: If4cfbd4ac0e63694e8144cb78c7fa34850647ff7
llvm-svn: 316072
Significantly reduces performancei (~30%) of gipfeli
(https://github.com/google/gipfeli)
I have not yet managed to reproduce this regression with the open-source
version of the benchmark on github, but will work with others to get a
reproducer to you later today.
llvm-svn: 315680
Summary:
If the pointer width is 32 bits and the calculated GEP offset is
negative, we call APInt::getLimitedValue(), which does a
*zero*-extension of the offset. That's wrong -- we should do an sext.
Fixes a bug introduced in rL314362 and found by Evgeny Astigeevich.
Reviewers: efriedma
Subscribers: sanjoy, javed.absar, llvm-commits, eastig
Differential Revision: https://reviews.llvm.org/D38557
llvm-svn: 314935
Function isLoweredToCall can only accept non-null function pointer, but a function pointer can be null for indirect function call. So check it before calling isLoweredToCall from getInstructionLatency.
Differential Revision: https://reviews.llvm.org/D38204
llvm-svn: 314927
Recommitting r314517 with the fix for handling ConstantExpr.
Original commit message:
Currently, getGEPCost() returns TCC_FREE whenever a GEP is a legal addressing
mode in the target. However, since it doesn't check its actual users, it will
return FREE even in cases where the GEP cannot be folded away as a part of
actual addressing mode. For example, if an user of the GEP is a call
instruction taking the GEP as a parameter, then the GEP may not be folded in
isel.
llvm-svn: 314923
Summary:
Currently, getGEPCost() returns TCC_FREE whenever a GEP is a legal addressing mode in the target.
However, since it doesn't check its actual users, it will return FREE even in cases
where the GEP cannot be folded away as a part of actual addressing mode.
For example, if an user of the GEP is a call instruction taking the GEP as a parameter,
then the GEP may not be folded in isel.
Reviewers: hfinkel, efriedma, mcrosier, jingyue, haicheng
Reviewed By: hfinkel
Subscribers: javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D38085
llvm-svn: 314517
Summary:
This avoids C++ UB if the GEP is weird and the calculation overflows
int64_t, and it's also observable in the cost model's results.
Such GEPs are almost surely not valid pointers, but LLVM nonetheless
generates them sometimes.
Reviewers: sanjoy
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38337
llvm-svn: 314362
Usually an intrinsic is a simple target instruction, it should have a small latency. A real function call has much larger latency. So handle the intrinsic call in function getInstructionLatency().
Differential Revision: https://reviews.llvm.org/D38104
llvm-svn: 314003
Static alloca usually doesn't generate any machine instructions, so it has 0 cost.
Differential Revision: https://reviews.llvm.org/D37879
llvm-svn: 313410
For instructions that unlikely generate machine instructions, they should also have 0 latency.
Differential Revision: https://reviews.llvm.org/D37833
llvm-svn: 313288
Current TargetTransformInfo can support throughput cost model and code size model, but sometimes we also need instruction latency cost model in different optimizations. Hal suggested we need a single public interface to query the different cost of an instruction. So I proposed following interface:
enum TargetCostKind {
TCK_RecipThroughput, ///< Reciprocal throughput.
TCK_Latency, ///< The latency of instruction.
TCK_CodeSize ///< Instruction code size.
};
int getInstructionCost(const Instruction *I, enum TargetCostKind kind) const;
All clients should mainly use this function to query the cost of an instruction, parameter <kind> specifies the desired cost model.
This patch also provides a simple default implementation of getInstructionLatency.
The default getInstructionLatency provides latency numbers for only small number of instruction classes, those latency numbers are only reasonable for modern OOO processors. It can be extended in following ways:
Add more detail into this function.
Add getXXXLatency function and call it from here.
Implement target specific getInstructionLatency function.
Differential Revision: https://reviews.llvm.org/D37170
llvm-svn: 312832
Summary:
Add patterns for
fptoui <16 x float> to <16 x i8>
fptoui <16 x float> to <16 x i16>
Reviewers: igorb, delena, craig.topper
Reviewed By: craig.topper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D37505
llvm-svn: 312704
Add missing SK_PermuteSingleSrc costs for AVX2 targets and earlier, also added some of the simpler SK_PermuteTwoSrc costs to support splitting of SK_PermuteSingleSrc shuffles
llvm-svn: 310632
This adds support for the new 32-bit vector float instructions of z14.
This includes:
- Enabling the instructions for the assembler/disassembler.
- CodeGen for the instructions, including new LLVM intrinsics.
- Scheduler description support for the instructions.
- Update to the vector cost function calculations.
In general, CodeGen support for the new v4f32 instructions closely
matches support for the existing v2f64 instructions.
llvm-svn: 308195
this patch updates the cost of addq\subq (add\subtract of vectors of 64bits)
based on the performance numbers of SLM arch.
Differential Revision: https://reviews.llvm.org/D33983
llvm-svn: 306974
The cost of an interleaved access was only implemented for AVX512. For other
X86 targets an overly conservative Base cost was returned, resulting in
avoiding vectorization where it is actually profitable to vectorize.
This patch starts to add costs for AVX2 for most prominent cases of
interleaved accesses (stride 3,4 chars, for now).
Note1: Improvements of up to ~4x were observed in some of EEMBC's rgb
workloads; There is also a known issue of 15-30% degradations on some of these
workloads, associated with an interleaved access followed by type
promotion/widening; the resulting shuffle sequence is currently inefficient and
will be improved by a series of patches that extend the X86InterleavedAccess pass
(such as D34601 and more to follow).
Note 2: The costs in this patch do not reflect port pressure penalties which can
be very dominant in the case of interleaved accesses since most of the shuffle
operations are restricted to a single port. Further tuning, that may incorporate
these considerations, will be done on top of the upcoming improved shuffle
sequences (that is, along with the abovementioned work to extend
X86InterleavedAccess pass).
Differential Revision: https://reviews.llvm.org/D34023
llvm-svn: 306238
The AArch64 instruction set has a few "widening" instructions (e.g., uaddl,
saddl, uaddw, etc.) that take one or more doubleword operands and produce
quadword results. The operands are automatically sign- or zero-extended as
appropriate. However, in LLVM IR, these extends are explicit. This patch
updates TTI to consider these widening instructions as single operations whose
cost is attached to the arithmetic instruction. It marks extends that are part
of a widening operation "free" and applies a sub-target specified overhead
(zero by default) to the arithmetic instructions.
Differential Revision: https://reviews.llvm.org/D32706
llvm-svn: 302582
Account for subvector extraction/insertion, helps prevent the vectorizers from selecting 256-bit vectors that will have to be split anyhow on AVX1 targets.
llvm-svn: 302378
Fixes PR31789 - When loop-vectorize tries to use these intrinsics for a
non-default address space pointer we fail with a "Calling a function with a
bad singature!" assertion. This patch solves this by adding the 'vector of
pointers' argument as an overloaded type which will determine the address
space.
Differential revision: https://reviews.llvm.org/D31490
llvm-svn: 302018
getArithmeticInstrCost(), getShuffleCost(), getCastInstrCost(),
getCmpSelInstrCost(), getVectorInstrCost(), getMemoryOpCost(),
getInterleavedMemoryOpCost() implemented.
Interleaved access vectorization enabled.
BasicTTIImpl::getCastInstrCost() improved to check for legal extending loads,
in which case the cost of the z/sext instruction becomes 0.
Review: Ulrich Weigand, Renato Golin.
https://reviews.llvm.org/D29631
llvm-svn: 300052
Currently the default C calling convention functions are treated
the same as compute kernels. Make this explicit so the default
calling convention can be changed to a non-kernel.
Converted with perl -pi -e 's/define void/define amdgpu_kernel void/'
on the relevant test directories (and undoing in one place that actually
wanted a non-kernel).
llvm-svn: 298444
getIntrinsicInstrCost() used to only compute scalarization cost based on types.
This patch improves this so that the actual arguments are checked when they are
available, in order to handle only unique non-constant operands.
Tests updates:
Analysis/CostModel/X86/arith-fp.ll
Transforms/LoopVectorize/AArch64/interleaved_cost.ll
Transforms/LoopVectorize/ARM/interleaved_cost.ll
The improvement in getOperandsScalarizationOverhead() to differentiate on
constants made it necessary to update the interleaved_cost.ll tests even
though they do not relate to intrinsics.
Review: Hal Finkel
https://reviews.llvm.org/D29540
llvm-svn: 297705
Newer ppc supports unaligned memory access, it reduces the cost of unaligned memory access significantly. This patch handles this case in PPCTTIImpl::getMemoryOpCost.
This patch fixes pr31492.
Differential Revision: https://reviews.llvm.org/D28630
This is resubmit of r292680, which was reverted by r293092. The internal application failures were actually caused by a source code bug.
llvm-svn: 295506
This reverts commit r292680. It is causing significantly worse
performance and test timeouts in our internal builds. I have already
routed reproduction instructions your way.
llvm-svn: 293092
Newer ppc supports unaligned memory access, it reduces the cost of unaligned memory access significantly. This patch handles this case in PPCTTIImpl::getMemoryOpCost.
This patch fixes pr31492.
Differential Revision: https://reviews.llvm.org/D28630
llvm-svn: 292680
updated instructions:
pmulld, pmullw, pmulhw, mulsd, mulps, mulpd, divss, divps, divsd, divpd, addpd and subpd.
special optimization case which replaces pmulld with pmullw\pmulhw\pshuf seq.
In case if the real operands bitwidth <= 16.
Differential Revision: https://reviews.llvm.org/D28104
llvm-svn: 291657
The original code considered only v2i64 as slow for this feature. This patch
consider all 128-bit long vector types as slow candidates.
In internal tests, extending this feature to all 128-bit vector types
resulted in an overall improvement of 1% on Exynos M1.
Differential revision: https://reviews.llvm.org/D27998
llvm-svn: 291616
The 'fast' costs should only work for shifts by uniform constants (uniform non-constant are lowered using the slow default implementation).
Logical shifts were not taking into account that we must mask the psrlw result, so the costs needed to be doubled.
Added missing AVX2/AVX512BW costs as well.
llvm-svn: 291391
X86 target does not provide any target specific cost calculation for interleave patterns.It uses the common target-independent calculation, which gives very high numbers. As a result, the scalar version is chosen in many cases. The situation on AVX-512 is even worse, since we have 3-src shuffles that significantly reduce the cost.
In this patch I calculate the cost on AVX-512. It will allow to compare interleave pattern with gather/scatter and choose a better solution (PR31426).
* Shiffle-broadcast cost will be changed in Simon's upcoming patch.
Differential Revision: https://reviews.llvm.org/D28118
llvm-svn: 290810
This patch checks that the SlowMisaligned128Store subtarget feature is set
when penalizing such stores in getMemoryOpCost.
Differential Revision: https://reviews.llvm.org/D27677
llvm-svn: 289845
VSX has instructions lxsiwax/lxsdx that can load 32/64 bit value into VSX register cheaply. That patch makes it known to memory cost model, so the vectorization of the test case in pr30990 is beneficial.
Differential Revision: https://reviews.llvm.org/D26713
llvm-svn: 288560
Currently when cost of scalar operations is evaluated the vector type is
used for scalar operations. Patch fixes this issue and fixes evaluation
of the vector operations cost.
Several test showed that vector cost model is too optimistic. It
allowed vectorization of 8 or less add/fadd operations, though scalar
code is faster. Actually, only for 16 or more operations vector code
provides better performance.
Differential Revision: https://reviews.llvm.org/D26277
llvm-svn: 288398
Summary:
This extends FCOPYSIGN support to 512-bit vectors.
I've also added tests to show what the 128-bit and 256-bit cases look like with broadcast loads.
Reviewers: delena, zvi, RKSimon, spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26791
llvm-svn: 287298
Add explicit v16i16/v32i8 ADD/SUB costs, matching the costs of v4i64/v8i32 - they were missing for some reason.
This has side effects on the LV max bandwidth tests (AVX1 now prefers 128-bit vectors vs AVX2 which still prefers 256-bit)
llvm-svn: 286832
This patch avoids scalarization of CTLZ by instead expanding to use CTPOP (ref: "Hacker's Delight") when the necessary operations are available.
This also adds the necessary cost models for X86 SSE2 targets (the main beneficiary) to ensure vectorization only happens when its useful.
Differential Revision: https://reviews.llvm.org/D25910
llvm-svn: 286233
There is a bug describing poor cost model for floating point operations:
Bug 29083 - [X86][SSE] Improve costs for floating point operations. This
patch is the second one in series of patches dealing with cost model.
Differential Revision: https://reviews.llvm.org/D25722
llvm-svn: 285564
With DQI but without VLX, lower v2i64 and v4i64 MUL operations with v8i64 MUL (vpmullq).
Updated cost table accordingly.
Differential Revision: https://reviews.llvm.org/D26011
llvm-svn: 285304
We were defaulting to SSE2 costs which weren't taking into account the availability of PBLENDW/PBLENDVB to improve merging of per-element shift results.
llvm-svn: 284939
We weren't accounting for legal types on every subtarget, meaning that many of the costs were using defaults.
We still don't correctly cost (or test) the 512-bit sdiv/udiv by uniform const cases, nor the power-of-2 cases.
llvm-svn: 284744
As discussed on PR28461 we currently miss the chance to lower "fptosi <2 x double> %arg to <2 x i32>" to cvttpd2dq due to its use of illegal types.
This patch adds support for fptosi to 2i32 from both 2f64 and 2f32.
It also recognises that cvttpd2dq zeroes the upper 64-bits of the xmm result (similar to D23797) - we still don't do this for the cvttpd2dq/cvttps2dq intrinsics - this can be done in a future patch.
Differential Revision: https://reviews.llvm.org/D23808
llvm-svn: 284459
This should fix:
https://llvm.org/bugs/show_bug.cgi?id=30433
There are a couple of open questions about the codegen:
1. Should we let scalar ops be scalars and avoid vector constant loads/splats?
2. Should we have a pass to combine constants such as the inverted pair that we have here?
Differential Revision: https://reviews.llvm.org/D25165
llvm-svn: 283119
Shifts with a uniform but non-constant count were considered very expensive to
vectorize, because the splat of the uniform count and the shift would tend to
appear in different blocks. That made the splat invisible to ISel, and we'd
scalarize the shift at codegen time.
Since r201655, CodeGenPrepare sinks those splats to be next to their use, and we
are able to select the appropriate vector shifts. This updates the cost model to
to take this into account by making shifts by a uniform cheap again.
Differential Revision: https://reviews.llvm.org/D23049
llvm-svn: 277782
This patch adds costs for the vectorized implementations of CTPOP, the default values were seriously underestimating the cost of these and was encouraging vectorization on targets where serialized use of POPCNT would be much better.
Differential Revision: https://reviews.llvm.org/D22456
llvm-svn: 276104
Make some AVX and AVX512 cast costs more precise.
Based on part of a patch by Elena Demikhovsky (D15604).
Differential Revision: http://reviews.llvm.org/D22064
llvm-svn: 275106
This is "cvtdq2ps" which does not appear to be particularly slow on any CPU
according to Agner's tables. Choosing "5" as a cost here as suggested in:
https://llvm.org/bugs/show_bug.cgi?id=21356
...but it seems very conservative given that the instruction is fully pipelined,
and I think these costs are supposed to model throughput.
Note that related costs are also most likely too high, but this fixes PR21356
and partly fixes PR28434.
llvm-svn: 274658
The cost model should not assume vector casts get completely scalarized, since
on targets that have vector support, the common case is a partial split up to
the legal vector size. So, when a vector cast gets split, the resulting casts
end up legal and cheap.
Instead of pessimistically assuming scalarization, base TTI can use the costs
the concrete TTI provides for the split vector, plus a fudge factor to account
for the cost of the split itself. This fudge factor is currently 1 by default,
except on AMDGPU where inserts and extracts are considered free.
Differential Revision: http://reviews.llvm.org/D21251
llvm-svn: 274642
This patch corresponds to review:
http://reviews.llvm.org/D20443
It changes the legalization strategy for illegal vector types from integer
promotion to widening. This only applies for vectors with elements of width
that is a multiple of a byte since we have hardware support for vectors with
1, 2, 3, 8 and 16 byte elements.
Integer promotion for vectors is quite expensive on PPC due to the sequence
of breaking apart the vector, extending the elements and reconstituting the
vector. Two of these operations are expensive.
This patch causes between minor and major improvements in performance on most
benchmarks. There are very few benchmarks whose performance regresses. These
regressions can be handled in a subsequent patch with a DAG combine (similar
to how this patch handles int -> fp conversions of illegal vector types).
llvm-svn: 274535
This is a resubmittion of 263158 change after fixing the existing problem with intrinsics mangling (see LTO and intrinsics mangling llvm-dev thread for details).
This patch fixes the problem which occurs when loop-vectorize tries to use @llvm.masked.load/store intrinsic for a non-default addrspace pointer. It fails with "Calling a function with a bad signature!" assertion in CallInst constructor because it tries to pass a non-default addrspace pointer to the pointer argument which has default addrspace.
The fix is to add pointer type as another overloaded type to @llvm.masked.load/store intrinsics.
Reviewed By: reames
Differential Revision: http://reviews.llvm.org/D17270
llvm-svn: 274043
This is a resubmittion of 263158 change after fixing the existing problem with intrinsics mangling (see LTO and intrinsics mangling llvm-dev thread for details).
This patch fixes the problem which occurs when loop-vectorize tries to use @llvm.masked.load/store intrinsic for a non-default addrspace pointer. It fails with "Calling a function with a bad signature!" assertion in CallInst constructor because it tries to pass a non-default addrspace pointer to the pointer argument which has default addrspace.
The fix is to add pointer type as another overloaded type to @llvm.masked.load/store intrinsics.
Reviewed By: reames
Differential Revision: http://reviews.llvm.org/D17270
llvm-svn: 273892
The BSWAP of vector types is quite efficiently implemented using vector shuffles on SSE/AVX targets, we should reflect the typical cost of this to encourage vectorization.
Differential Revision: http://reviews.llvm.org/D21521
llvm-svn: 273217
The costs are somewhat hand-wavy, but should be much closer to the truth
than what we get from BasicTTI.
Differential Revision: http://reviews.llvm.org/D21156
llvm-svn: 272406
As discussed on PR24888, until SSE42 we don't have access to PCMPGTQ for v2i64 comparisons, but the cost models don't reflect this, resulting in over-optimistic vectorizaton.
This patch adds SSE2 'base level' costs that match what a typical target is capable of and only reduces the v2i64 costs at SSE42.
Technically SSE41 provides a PCMPEQQ v2i64 equality test, but as getCmpSelInstrCost doesn't give us a way to discriminate between comparison test types we can't easily make use of this, otherwise we could split the cost of integer equality and greater-than tests to give better costings of each.
Differential Revision: http://reviews.llvm.org/D20057
llvm-svn: 268972
Summary:
rL256194 transforms truncations between vectors of integers into PACKUS/PACKSS
operations during DAG combine. This generates better code for truncate, so cost
of truncate needs to be changed but looks like it got changed only in SSE2 table
Whereas this change is also applicable for SSE4.1, so the cost of truncate needs
to be changed for that as well. Cost of “TRUNCATE v16i32 to v16i8” & “TRUNCATE
v16i16 to v16i8” should be same in SSE4.1 & SSE2 table. Removing their cost from
SSE4.1, so it will fall back to SSE2.
Reviewers: Simon Pilgrim
llvm-svn: 267123
This is a resubmittion of 263158 change.
This patch fixes the problem which occurs when loop-vectorize tries to use @llvm.masked.load/store intrinsic for a non-default addrspace pointer. It fails with "Calling a function with a bad signature!" assertion in CallInst constructor because it tries to pass a non-default addrspace pointer to the pointer argument which has default addrspace.
The fix is to add pointer type as another overloaded type to @llvm.masked.load/store intrinsics.
Reviewed By: reames
Differential Revision: http://reviews.llvm.org/D17270
llvm-svn: 266086
PPC has a vector popcount, this lets the vectorizer use the correct cost
for it. Tweak X86 test to use an intrinsic that's actually scalarized (we
have a somewhat efficient lowering for vector popcount using SSE, the
cost model finds that now).
llvm-svn: 265005
This patch fixes the problem which occurs when loop-vectorize tries to use @llvm.masked.load/store intrinsic for a non-default addrspace pointer. It fails with "Calling a function with a bad signature!" assertion in CallInst constructor because it tries to pass a non-default addrspace pointer to the pointer argument which has default addrspace.
The fix is to add pointer type as another overloaded type to @llvm.masked.load/store intrinsics.
Reviewed By: reames
Differential Revision: http://reviews.llvm.org/D17270
llvm-svn: 263158
The cost is calculated for all X86 targets. When gather/scatter instruction
is not supported we calculate the cost of scalar sequence.
Differential revision: http://reviews.llvm.org/D15677
llvm-svn: 256519
This patch transforms truncation between vectors of integers into
X86ISD::PACKUS/PACKSS operations during DAG combine. We don't do it in
lowering phase because after type legalization, the original truncation
will be turned into a BUILD_VECTOR with each element that is extracted
from a vector and then truncated, and from them it is difficult to do
this optimization. This greatly improves the performance of truncations
on some specific types.
Cost table is updated accordingly.
Differential revision: http://reviews.llvm.org/D14588
llvm-svn: 256194
Previously in the conversion cost table there are no entries for integer-integer
conversions on SSE2. This will result in imprecise costs for certain vectorized
operations. This patch adds those entries for SSE2 and SSE4.1. The cost numbers
are counted from the result of running llc on the new test case in this patch.
Differential revision: http://reviews.llvm.org/D15132
llvm-svn: 255315
Currently in LLVM's cost model, a vectorized arithmetic instruction will have
high cost if its type is split into multiple registers. However, this
punishment is too heavy and unnecessary. The overhead of the split should not
be on arithmetic instructions but instructions that implement the split. Note
that during vectorization we have calculated the register pressure, and we
only choose proper interleaving factor (and also vectorization factor) so
that we don't use more registers than the maximum number.
Here is a very simple example: if a vadd has the cost 1, and if we double VF
so that we need two registers to perform it, then its cost will become 4 with
the current implementation, which will prevent us to use larger VF.
Differential revision: http://reviews.llvm.org/D15159
llvm-svn: 254671
I checked and updated the cost of AVX-512 conversion operations. Added cost of conversion operations in DQ mode.
Conversion of illegal types that requires vector split is not calculated right now (like for other X86 targets).
Differential Revision: http://reviews.llvm.org/D15074
llvm-svn: 254494
The cost for scalarized operations is computed as N * (scalar operation
cost + 1 extractelement + 1 insertelement). This partially fixes
inflating the cost of scalarized operations since every operation is
scalarized and free. I don't think we want any cost asociated with
scalarization, but for now insertelement is still counted. I'm not sure
if we should pretend that insertelement is also free, or add a way
to compute a custom scalarization cost.
llvm-svn: 254438
Cost calculation for vector GEP failed with due to invalid cast to GEP index operand.
The bug is fixed, added a test.
http://reviews.llvm.org/D14976
llvm-svn: 254408
The underlying issues surrounding codegen for 32-bit vselects have been resolved. The pessimistic costs for 64-bit vselects remain due to the bad
scalarization that is still happening there.
I tested this on A57 in T32, A32 and A64 modes. I saw no regressions, and some improvements.
From my benchmarks, I saw these improvements in A57 (T32)
spec.cpu2000.ref.177_mesa 5.95%
lnt.SingleSource/Benchmarks/Shootout/strcat 12.93%
lnt.MultiSource/Benchmarks/MiBench/telecomm-CRC32/telecomm-CRC32 11.89%
I also measured A57 A32, A53 T32 and A9 T32 and found no performance regressions. I see much bigger wins in third-party benchmarks with this change
Differential Revision: http://reviews.llvm.org/D14743
llvm-svn: 253349
The XOP shifts just have logical/arithmetic versions and the left/right shifts are controlled by whether the value is positive/negative. Because of this I've added new X86ISD nodes instead of trying to force them to use the existing shift nodes.
Additionally Excavator cores (bdver4) support XOP and AVX2 - meaning that it should use the AVX2 shifts when it can and fall back to XOP in other cases.
Differential Revision: http://reviews.llvm.org/D8690
llvm-svn: 248878
Summary:
We are not scalarizing the wide selects in codegen for i16 and i32 and
therefore we can remove the amortization factor. We still have issues
with i64 vectors in codegen though.
Reviewers: mcrosier
Subscribers: mcrosier, aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D12724
llvm-svn: 247156
Pre-P8, when we generate code for unaligned vector loads (for Altivec and QPX
types), even when accounting for the combining that takes place for multiple
consecutive such loads, there is at least one load instructions and one
permutation for each load. Make sure the cost reported reflects the cost of the
permutes as well.
llvm-svn: 246807
I'm adding a regression test to better cover code generation for unaligned
vector loads and stores, but there's no functional change to the code
generation here. There is an improvement to the cost model for unaligned vector
loads and stores, mostly for QPX (for which we were not previously accounting
for the permutation-based loads), and the cost model implementation is cleaner.
llvm-svn: 246712
Summary:
This change limits the minimum cost of an insert/extract
element operation to 2 in cases where this would result
in mixing of NEON and VFP code.
Reviewers: rengolin
Subscribers: mssimpso, aemerson, llvm-commits, rengolin
Differential Revision: http://reviews.llvm.org/D12030
llvm-svn: 245225
This patch vectorizes the v2i64/v4i64 ASHR shift operations - the last remaining integer vector shifts that are still being transferred to/from the scalar unit to be completed.
Differential Revision: http://reviews.llvm.org/D11439
llvm-svn: 243569
r243250 appeared to break clang/test/Analysis/dead-store.c on one of the build
slaves, but I couldn't reproduce this failure locally. Probably a false
positive as I saw this test was broken by r243246 or r243247 too but passed
later without people fixing anything.
llvm-svn: 243253
Summary:
This patch updates TargetTransformInfoImplCRTPBase::getGEPCost to consider
addressing modes. It now returns TCC_Free when the GEP can be completely folded
to an addresing mode.
I started this patch as I refactored SLSR. Function isGEPFoldable looks common
and is indeed used by some WIP of mine. So I extracted that logic to getGEPCost.
Furthermore, I noticed getGEPCost wasn't directly tested anywhere. The best
testing bed seems CostModel, but its getInstructionCost method invokes
getAddressComputationCost for GEPs which provides very coarse estimation. So
this patch also makes getInstructionCost call the updated getGEPCost for GEPs.
This change inevitably breaks some tests because the cost model changes, but
nothing looks seriously wrong -- if we believe the new cost model is the right
way to go, these tests should be updated.
This patch is not perfect yet -- the comments in some tests need to be updated.
I want to know whether this is a right approach before fixing those details.
Reviewers: chandlerc, hfinkel
Subscribers: aschwaighofer, llvm-commits, aemerson
Differential Revision: http://reviews.llvm.org/D9819
llvm-svn: 243250
While the v4i32 shl operation is already vectorized using a cvttps2dq/pmulld pattern, the lshr/ashr opeations are still scalarized.
This patch adds vectorization support for non-uniform v4i32 shift operations - it splats constant shift amounts to allow them to use the immediate sse shift instructions, or extracts/zero-extends non-constant shift amounts. The individual results are then blended together.
Differential Revision: http://reviews.llvm.org/D11063
llvm-svn: 241989
This patch adds vectorization support for uniform constant i64 arithmetic shift right operators.
Differential Revision: http://reviews.llvm.org/D9645
llvm-svn: 241514
Merged separate (but equivalent) SSE2/AVX512F tests.
Removed codegen tests since these are already done better in test/CodeGen/X86.
The actual cost values still need to be updated to match recent codegen improvements.
llvm-svn: 240219
This patch ensures that SHL/SRL/SRA shifts for i8 and i16 vectors avoid scalarization. It builds on the existing i8 SHL vectorized implementation of moving the shift bits up to the sign bit position and separating the 4, 2 & 1 bit shifts with several improvements:
1 - SSE41 targets can use (v)pblendvb directly with the sign bit instead of performing a comparison to feed into a VSELECT node.
2 - pre-SSE41 targets were masking + comparing with an 0x80 constant - we avoid this by using the fact that a set sign bit means a negative integer which can be compared against zero to then feed into VSELECT, avoiding the need for a constant mask (zero generation is much cheaper).
3 - SRA i8 needs to be unpacked to the upper byte of a i16 so that the i16 psraw instruction can be correctly used for sign extension - we have to do more work than for SHL/SRL but perf tests indicate that this is still beneficial.
The i16 implementation is similar but simpler than for i8 - we have to do 8, 4, 2 & 1 bit shifts but less shift masking is involved. SSE41 use of (v)pblendvb requires that the i16 shift amount is splatted to both bytes however.
Tested on SSE2, SSE41 and AVX machines.
Differential Revision: http://reviews.llvm.org/D9474
llvm-svn: 239509
Essentially the same as the GEP change in r230786.
A similar migration script can be used to update test cases, though a few more
test case improvements/changes were required this time around: (r229269-r229278)
import fileinput
import sys
import re
pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)")
for line in sys.stdin:
sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line))
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7649
llvm-svn: 230794
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
First, don't combine bit masking into vector shuffles (even ones the
target can handle) once operation legalization has taken place. Custom
legalization of vector shuffles may exist for these patterns (making the
predicate return true) but that custom legalization may in some cases
produce the exact bit math this matches. We only really want to handle
this prior to operation legalization.
However, the x86 backend, in a fit of awesome, relied on this. What it
would do is mark VSELECTs as expand, which would turn them into
arithmetic, which this would then match back into vector shuffles, which
we would then lower properly. Amazing.
Instead, the second change is to teach the x86 backend to directly form
vector shuffles from VSELECT nodes with constant conditions, and to mark
all of the vector types we support lowering blends as shuffles as custom
VSELECT lowering. We still mark the forms which actually support
variable blends as *legal* so that the custom lowering is bypassed, and
the legal lowering can even be used by the vector shuffle legalization
(yes, i know, this is confusing. but that's how the patterns are
written).
This makes the VSELECT lowering much more sensible, and in fact should
fix a bunch of bugs with it. However, as you'll see in the test cases,
right now what it does is point out the *hilarious* deficiency of the
new vector shuffle lowering when it comes to blends. Fortunately, my
very next patch fixes that. I can't submit it yet, because that patch,
somewhat obviously, forms the exact and/or pattern that the DAG combine
is matching here! Without this patch, teaching the vector shuffle
lowering to produce the right code infloops in the DAG combiner. With
this patch alone, we produce terrible code but at least lower through
the right paths. With both patches, all the regressions here should be
fixed, and a bunch of the improvements (like using 2 shufps with no
memory loads instead of 2 andps with memory loads and an orps) will
stay. Win!
There is one other change worth noting here. We had hilariously wrong
vectorization cost estimates for vselect because we fell through to the
code path that assumed all "expand" vector operations are scalarized.
However, the "expand" lowering of VSELECT is vector bit math, most
definitely not scalarized. So now we go back to the correct if horribly
naive cost of "1" for "not scalarized". If anyone wants to add actual
modeling of shuffle costs, that would be cool, but this seems an
improvement on its own. Note the removal of 16 and 32 "costs" for doing
a blend. Even in SSE2 we can blend in fewer than 16 instructions. ;] Of
course, we don't right now because of OMG bad code, but I'm going to fix
that. Next patch. I promise.
llvm-svn: 229835
type erased interface and a single analysis pass rather than an
extremely complex analysis group.
The end result is that the TTI analysis can contain a type erased
implementation that supports the polymorphic TTI interface. We can build
one from a target-specific implementation or from a dummy one in the IR.
I've also factored all of the code into "mix-in"-able base classes,
including CRTP base classes to facilitate calling back up to the most
specialized form when delegating horizontally across the surface. These
aren't as clean as I would like and I'm planning to work on cleaning
some of this up, but I wanted to start by putting into the right form.
There are a number of reasons for this change, and this particular
design. The first and foremost reason is that an analysis group is
complete overkill, and the chaining delegation strategy was so opaque,
confusing, and high overhead that TTI was suffering greatly for it.
Several of the TTI functions had failed to be implemented in all places
because of the chaining-based delegation making there be no checking of
this. A few other functions were implemented with incorrect delegation.
The message to me was very clear working on this -- the delegation and
analysis group structure was too confusing to be useful here.
The other reason of course is that this is *much* more natural fit for
the new pass manager. This will lay the ground work for a type-erased
per-function info object that can look up the correct subtarget and even
cache it.
Yet another benefit is that this will significantly simplify the
interaction of the pass managers and the TargetMachine. See the future
work below.
The downside of this change is that it is very, very verbose. I'm going
to work to improve that, but it is somewhat an implementation necessity
in C++ to do type erasure. =/ I discussed this design really extensively
with Eric and Hal prior to going down this path, and afterward showed
them the result. No one was really thrilled with it, but there doesn't
seem to be a substantially better alternative. Using a base class and
virtual method dispatch would make the code much shorter, but as
discussed in the update to the programmer's manual and elsewhere,
a polymorphic interface feels like the more principled approach even if
this is perhaps the least compelling example of it. ;]
Ultimately, there is still a lot more to be done here, but this was the
huge chunk that I couldn't really split things out of because this was
the interface change to TTI. I've tried to minimize all the other parts
of this. The follow up work should include at least:
1) Improving the TargetMachine interface by having it directly return
a TTI object. Because we have a non-pass object with value semantics
and an internal type erasure mechanism, we can narrow the interface
of the TargetMachine to *just* do what we need: build and return
a TTI object that we can then insert into the pass pipeline.
2) Make the TTI object be fully specialized for a particular function.
This will include splitting off a minimal form of it which is
sufficient for the inliner and the old pass manager.
3) Add a new pass manager analysis which produces TTI objects from the
target machine for each function. This may actually be done as part
of #2 in order to use the new analysis to implement #2.
4) Work on narrowing the API between TTI and the targets so that it is
easier to understand and less verbose to type erase.
5) Work on narrowing the API between TTI and its clients so that it is
easier to understand and less verbose to forward.
6) Try to improve the CRTP-based delegation. I feel like this code is
just a bit messy and exacerbating the complexity of implementing
the TTI in each target.
Many thanks to Eric and Hal for their help here. I ended up blocked on
this somewhat more abruptly than I expected, and so I appreciate getting
it sorted out very quickly.
Differential Revision: http://reviews.llvm.org/D7293
llvm-svn: 227669
AVX2 is available.
According to IACA, the new lowering has a throughput of 8 cycles instead of 13
with the previous one.
Althought this lowering kicks in some SPECs benchmarks, the performance
improvement was within the noise.
Correctness testing has been done for the whole range of uint32_t with the
following program:
uint4 v = (uint4) {0,1,2,3};
uint32_t i;
//Check correctness over entire range for uint4 -> float4 conversion
for( i = 0; i < 1U << (32-2); i++ )
{
float4 t = test(v);
float4 c = correct(v);
if( 0xf != _mm_movemask_ps( t == c ))
{
printf( "Error @ %vx: %vf vs. %vf\n", v, c, t);
return -1;
}
v += 4;
}
Where "correct" is the old lowering and "test" the new one.
The patch adds a test case for the two custom lowering instruction.
It also modifies the vector cost model, which is why cast.ll and uitofp.ll are
modified.
2009-02-26-MachineLICMBug.ll is also modified because we now hoist 7
instructions instead of 4 (3 more constant loads).
rdar://problem/18153096>
llvm-svn: 221657
The default implementation of getCmpSelInstrCost, which provides the cost of
icmp/fcmp/select instructions, did not deal sensibly with illegal vector types
that were scalarized. We'd ask for the legalization cost of the vector type,
which would return something like (4, f64) given an input of <4 x double>, and
we'd then check the TLI status of the ISD opcode on that scalar type. This would
result in querying (ISD::VSELECT, f64), for example. Amusingly enough,
ISD::VSELECT on scalar types is marked as Legal by default (as with most other
operations), and most backends never change this because VSELECT is never
generated on scalars. However, seeing the resulting operation as Legal, we'd
neglect to add the scalarization cost before returning. The result is that we'd
grossly under-estimate the cost of cmps/selects on illegal vector types.
Now, if type legalization clearly results in scalarization, we skip the early
return and add the scalarization cost.
llvm-svn: 217859
Cross-class copies being expensive is actually a trait of the microarchitecture, but as I haven't yet seen an example of a microarchitecture where they're cheap it seems best to just enable this by default, covering the non-mcpu build case.
llvm-svn: 217674
This patch:
1) Improves the cost model for x86 alternate shuffles (originally
added at revision 211339);
2) Teaches the Cost Model Analysis pass how to analyze alternate shuffles.
Alternate shuffles are a special kind of blend; on x86, we can often
easily lowered alternate shuffled into single blend
instruction (depending on the subtarget features).
The existing cost model didn't take into account subtarget features.
Also, it had a couple of "dead" entries for vector types that are never
legal (example: on x86 types v2i32 and v2f32 are not legal; those are
always either promoted or widened to 128-bit vector types).
The new x86 cost model takes into account what target features we have
before returning the shuffle cost (i.e. the number of instructions
after the blend is lowered/expanded).
This patch also teaches the Cost Model Analysis how to identify and analyze
alternate shuffles (i.e. 'SK_Alternate' shufflevector instructions):
- added function 'isAlternateVectorMask';
- added some logic to check if an instruction is a alternate shuffle and, in
case, call the target specific TTI to get the corresponding shuffle cost;
- added a test to verify the cost model analysis on alternate shuffles.
llvm-svn: 212296
This commit starts with a "git mv ARM64 AArch64" and continues out
from there, renaming the C++ classes, intrinsics, and other
target-local objects for consistency.
"ARM64" test directories are also moved, and tests that began their
life in ARM64 use an arm64 triple, those from AArch64 use an aarch64
triple. Both should be equivalent though.
This finishes the AArch64 merge, and everyone should feel free to
continue committing as normal now.
llvm-svn: 209577
BasicTTI::getMemoryOpCost must explicitly check for non-simple types; setting
AllowUnknown=true with TLI->getSimpleValueType is not sufficient because, for
example, non-power-of-two vector types return non-simple EVTs (not MVT::Other).
llvm-svn: 206150
This provides more realistic costs for the insert/extractelement instructions
(which are load/store pairs), accounts for the cheap unaligned Altivec load
sequence, and for unaligned VSX load/stores.
Bad news:
MultiSource/Applications/sgefa/sgefa - 35% slowdown (this will require more investigation)
SingleSource/Benchmarks/McGill/queens - 20% slowdown (we no longer vectorize this, but it was a constant store that was scalarized)
MultiSource/Benchmarks/FreeBench/pcompress2/pcompress2 - 2% slowdown
Good news:
SingleSource/Benchmarks/Shootout/ary3 - 54% speedup
SingleSource/Benchmarks/Shootout-C++/ary - 40% speedup
MultiSource/Benchmarks/Ptrdist/ks/ks - 35% speedup
MultiSource/Benchmarks/FreeBench/neural/neural - 30% speedup
MultiSource/Benchmarks/TSVC/Symbolics-flt/Symbolics-flt - 20% speedup
Unfortunately, estimating the costs of the stack-based scalarization sequences
is hard, and adjusting these costs is like a game of whac-a-mole :( I'll
revisit this again after we have better codegen for vector extloads and
truncstores and unaligned load/stores.
llvm-svn: 205658
When a vector type legalizes to a larger vector type, and the target does not
support the associated extending load (or truncating store), then legalization
will scalarize the load (or store) resulting in an associated scalarization
cost. BasicTTI::getMemoryOpCost needs to account for this.
Between this, and r205487, PowerPC on the P7 with VSX enabled shows:
MultiSource/Benchmarks/PAQ8p/paq8p: 43% speedup
SingleSource/Benchmarks/BenchmarkGame/puzzle: 51% speedup
SingleSource/UnitTests/Vectorizer/gcc-loops 28% speedup
(some of these are new; some of these, such as PAQ8p, just reverse regressions
that VSX support would trigger)
llvm-svn: 205495
For an cast (extension, etc.), the currently logic predicts a low cost if the
associated operation (keyed on the destination type) is legal (or promoted).
This is not true when the number of values required to legalize the type is
changing. For example, <8 x i16> being sign extended by <8 x i32> is not
generically cheap on PPC with VSX, even though sign extension to v4i32 is
legal, because two output v4i32 values are required compared to the single
v8i16 input value, and without custom logic in the target, this conversion will
scalarize.
llvm-svn: 205487
This adds a second implementation of the AArch64 architecture to LLVM,
accessible in parallel via the "arm64" triple. The plan over the
coming weeks & months is to merge the two into a single backend,
during which time thorough code review should naturally occur.
Everything will be easier with the target in-tree though, hence this
commit.
llvm-svn: 205090
the legalization cost must be included to get an accurate
estimation of the total cost of the scalarized vector.
The inaccurate cost triggered unprofitable SLP vectorization on
32-bit X86.
Summary:
Include legalization overhead when computing scalarization cost
Reviewers: hfinkel, nadav
CC: chandlerc, rnk, llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D2992
llvm-svn: 203509
'OK_NonUniformConstValue' to identify operands which are constants but
not constant splats.
The cost model now allows returning 'OK_NonUniformConstValue'
for non splat operands that are instances of ConstantVector or
ConstantDataVector.
With this change, targets are now able to compute different costs
for instructions with non-uniform constant operands.
For example, On X86 the cost of a vector shift may vary depending on whether
the second operand is a uniform or non-uniform constant.
This patch applies the following changes:
- The cost model computation now takes into account non-uniform constants;
- The cost of vector shift instructions has been improved in
X86TargetTransformInfo analysis pass;
- BBVectorize, SLPVectorizer and LoopVectorize now know how to distinguish
between non-uniform and uniform constant operands.
Added a new test to verify that the output of opt
'-cost-model -analyze' is valid in the following configurations: SSE2,
SSE4.1, AVX, AVX2.
llvm-svn: 201272
The most important part of this is probably adding any cost at all for
operations like zext <8 x i8> to <8 x i32>. Before they were being
recorded as extremely costly (24, I believe) which made LLVM fall back
on a 4-wide vectorisation of a loop.
It also rebalances the values for sext, zext and trunc. Lacking any
other sane metric that might work across CPU microarchitectures I went
for instructions. This seems to be in reasonable accord with the rest
of the table (sitofp, ...) though no doubt at least one value is
sub-optimal for some bizarre reason.
Finally, separate AVX and AVX2 values are provided where appropriate.
The CodeGen is quite different in many cases.
rdar://problem/15981990
llvm-svn: 200928
Upcoming SLP vectorization improvements will want to be able to estimate costs
of horizontal reductions. Add infrastructure to support this.
We model reductions as a series of (shufflevector,add) tuples ultimately
followed by an extractelement. For example, for an add-reduction of <4 x float>
we could generate the following sequence:
(v0, v1, v2, v3)
\ \ / /
\ \ /
+ +
(v0+v2, v1+v3, undef, undef)
\ /
((v0+v2) + (v1+v3), undef, undef)
%rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef,
<4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
%bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
%rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef,
<4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
%r = extractelement <4 x float> %bin.rdx8, i32 0
This commit adds a cost model interface "getReductionCost(Opcode, Ty, Pairwise)"
that will allow clients to ask for the cost of such a reduction (as backends
might generate more efficient code than the cost of the individual instructions
summed up). This interface is excercised by the CostModel analysis pass which
looks for reduction patterns like the one above - starting at extractelements -
and if it sees a matching sequence will call the cost model interface.
We will also support a second form of pairwise reduction that is well supported
on common architectures (haddps, vpadd, faddp).
(v0, v1, v2, v3)
\ / \ /
(v0+v1, v2+v3, undef, undef)
\ /
((v0+v1)+(v2+v3), undef, undef, undef)
%rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
<4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
%rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
<4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
%bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
%rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
<4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
%rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
<4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
%bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
%r = extractelement <4 x float> %bin.rdx.1, i32 0
llvm-svn: 190876
- Instead of setting the suffixes in a bunch of places, just set one master
list in the top-level config. We now only modify the suffix list in a few
suites that have one particular unique suffix (.ml, .mc, .yaml, .td, .py).
- Aside from removing the need for a bunch of lit.local.cfg files, this enables
4 tests that were inadvertently being skipped (one in
Transforms/BranchFolding, a .s file each in DebugInfo/AArch64 and
CodeGen/PowerPC, and one in CodeGen/SI which is now failing and has been
XFAILED).
- This commit also fixes a bunch of config files to use config.root instead of
older copy-pasted code.
llvm-svn: 188513
This patch fixes the multiple breakages on ARM test-suite after the SLP
vectorizer was introduced by default on O3. The problem was an illegal
vector type on ARMTTI::getCmpSelInstrCost() <3 x i1> which is not simple.
The guard protects this code from breaking (cause of the problems) but
doesn't fix the issue that is generating the odd vector in the first
place, which also needs to be investigated.
llvm-svn: 187658
This fixes an oversight that Intrinsic::nearbyint was not being mapped to
ISD::FNEARBYINT (thus fixing the over-optimistic cost we were assigning to
nearbyint calls for some targets).
llvm-svn: 185783
Evandro Menezes