There was no pattern to fold into these instructions. This patch adds
the pattern obtained from the following ACLE intrinsics so that they
generate sqdmlal/sqdmlsl instructions instead of separate sqdmull and
sqadd/sqsub instructions:
- vqdmlalh_s16, vqdmlslh_s16
- vqdmlalh_lane_s16, vqdmlalh_laneq_s16, vqdmlslh_lane_s16,
vqdmlslh_laneq_s16 (when the lane index is 0)
It also modifies the result of the existing pattern for the latter, when
the lane index is not 0, to use the v1i32_indexed instructions instead
of the v4i16_indexed ones.
Fixes#49997.
Differential Revision: https://reviews.llvm.org/D131700
Currenlty all temporal loads are mapped to `LDP` or `LDR`. This patch will map all the non temporal 256-bit loads into `LDNP`. Future patches should address other non-temporal loads.
Reviewed By: fhahn, dmgreen
Differential Revision: https://reviews.llvm.org/D131773
This reverts commit 38c2366b3f.
This patch seems to break boostraping LLVM with `-fglobal-isel -O3`
on AArch64 hardware. Without the revert, there are 500+ test
failures for the `check-llvm-codegen-x86` target.
A bfloat select operation will currently crash, but is allowed from C.
This adds handling for the operation, turning it into a FCSELHrrr if
fullfp16 is present, or converting it to a FCSELSrrr if not. The
FCSELSrrr is created via using INSERT_SUBREG/EXTRACT_SUBREG to convert
the bf16 to a f32 and using the f32 pattern for FCSELSrrr. (I originally
attempted to do this via a tablegen pattern, but it appears that the
nzcv glue is places onto the wrong node, causing it to be forgotten and
incorrect scheduling to be emitted).
The FCSELSrrr can also be used for fp16 selects when +fullfp16 is not
present, which helps avoid an unnecessary promotion to f32.
Differential Revision: https://reviews.llvm.org/D131253
Currently fcopysign for VLS vectors lowers through NEON even when the
vector width is wider than a NEON vector, causing bad codegen as the
vectors are split. This patch causes SVE to be used for these vectors
instead, giving much better codegen on wide VLS vectors.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D128642
After D121595 was commited, I noticed regressions assosicated with small trip
count numbersvectorisation by tail folding with scalable vectors. As a solution
for those issues I propose to introduce the minimal trip count threshold value.
Differential Revision: https://reviews.llvm.org/D130755
This patch adds the names of the Arm Architecture Reference Manual (ARM)
features to the corresponding Subtarget Features in the AArch64 backend
and target parser.
The aim of this is to make it clearer what architectural features a
subtarget feature might enable (so, which features a CPU must provide to
support that subtarget feature), and so make it easier to add new CPUs
in the future.
Differential Revision: https://reviews.llvm.org/D131257
The register operand of DBG_VALUE is not selected to a proper register
bank in both AArch64 and X86. This would cause getRegClass crash after
global ISel. After discussion, we think the MIR should assume all
vritual register should be set proper register class after global ISel,
so this patch is to fix the gap of DBG_VALUE for AArch64 and X86.
Differential Revision: https://reviews.llvm.org/D129037
1. Missing instruction information (FTSSEL, FMSB, PFIRST and RDFFR)
is added and CompleteModel is set to one.
2. Information for pseudo SVE instructions is added. Those
instructions are present at the time of scheduling.
3. Resource and latency information for SVE instructions is modified
to be more accurate.
For example, the description for CMPEQ, which consumes one cycle
each of unit FLA and PPR, is as follows.
```
Previous:
def A64FXGI01 : ProcResGroup<[A64FXIPFLA, A64FXIPPR]>;
def A64FXWrite_4Cyc_GI01 : SchedWriteRes<[A64FXGI01]> {...
Modified:
def A64FXGI0 : ProcResGroup<[A64FXIPFLA]>;
def A64FXGI1 : ProcResGroup<[A64FXIPPR]>;
def A64FXWrite_CMP : SchedWriteRes<[A64FXGI0, A64FXGI1]> {...
```
Reference: A64FX Microarchitecture Manual (Table 16-3)
https://github.com/fujitsu/A64FX/blob/master/doc/A64FX_Microarchitecture_Manual_en_1.7.pdf
Reviewed By: dmgreen, kawashima-fj
Differential Revision: https://reviews.llvm.org/D131165
Add patterns to select predicated instructions when lowering:
fadd(a, select(mask, b, splat(0)))
fsub(a, select(mask, b, splat(0)))
'fadd' is unsafe unless no-signed zeros fast-math flag is set, since
-0.0 + 0.0 = 0.0
changes the sign. Alive2: https://alive2.llvm.org/ce/z/wbhJh_
Also adds FMA patterns for:
fadd(a, select(mask, mul(b, c), splat(0))) -> fmla(a, mask, b, c)
fsub(a, select(mask, mul(b, c), splat(0))) -> fmla(a, mask, b, c)
These patterns require the 'contract' fast-math flag to be set, and the
fadd 'nsz' as above.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D130564
NOTE: i8 vector splats are ignored because the immediate range of
DUP already has full coverage.
Differential Revision: https://reviews.llvm.org/D131078
This is a simple addition to emitConditionalComparison, to match CCMP
with immediates using getIConstantVRegValWithLookThrough, letting it
select the CCMPri variants of the instructions.
Differential Revision: https://reviews.llvm.org/D131073
1) Overloaded (instruction-based) method is a wrapper around the current (opcode-based) method.
2) This patch also changes a few callsites (VectorCombine.cpp,
SLPVectorizer.cpp, CodeGenPrepare.cpp) to call the overloaded method.
3) This is a split of D128302.
Differential Revision: https://reviews.llvm.org/D131114
This patch ensures consistency in the construction of FP_ROUND nodes
such that they always use ISD::TargetConstant instead of ISD::Constant.
This additionally fixes a bug in the AArch64 SVE backend where patterns
were matching against TargetConstant nodes and sometimes failing when
passed a Constant node.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D130370
rGcf97e0ec42b8 makes $x18 to be treated as callee-saved in functions with
Windows calling convention on non-Windows OSes.
Here we mark $x18 as callee-saved for functions with Windows calling
convention on Darwin, as well as on other non-Windows platforms, in
order to prevent some miscompilations (like miscompilation of
win64cc-darwin-backup-x18.ll).
Since getCalleeSavedRegs doesn't return x18 in list of callee-saved
registers, assignCalleeSavedSpillSlots and determineCalleeSaves
consider different sets of registers as callee-saved. It causes an
error:
```
Assertion failed: ((!HasCalleeSavedStackSize || getCalleeSavedStackSize() == Size) && "Invalid size calculated for callee saves"), function getCalleeSavedStackSize, file
AArch64MachineFunctionInfo.h, line 292.
```
Differential Revision: https://reviews.llvm.org/D130676
This folds a v4i32 Mul(And(Srl(X, 15), 0x10001), 0xffff) into a v8i16
CMLTz instruction. The Srl and And extract the top bit (whether the
input is negative) and the Mul sets all values in the i16 half to all
1/0 depending on if that top bit was set. This is equivalent to a v8i16
CMLTz instruction. The same applies to other sizes with equivalent
constants.
Differential Revision: https://reviews.llvm.org/D130874
If we have interleave groups in the loop we want to vectorise then
we should fall back on normal vectorisation with a scalar epilogue. In
such cases when tail-folding is enabled we'll almost certainly go on to
create vplans with very high costs for all vector VFs and fall back on
VF=1 anyway. This is likely to be worse than if we'd just used an
unpredicated vector loop in the first place.
Once the vectoriser has proper support for analysing all the costs
for each combination of VF and vectorisation style, then we should
be able to remove this.
Added an extra test here:
Transforms/LoopVectorize/AArch64/sve-tail-folding-option.ll
Differential Revision: https://reviews.llvm.org/D128342
2xi64 is the legalized type for wide reductions (like 16xi64) and setting the
cost to 2 makes `load-reduce` and `load-zext-reduce` patterns profitable.
The few performance measurments that I did on an aarch64 machine confirm that
these patterns are actually faster when vectorized.
Differential Revision: https://reviews.llvm.org/D130740
A build vector of two extracted elements is equivalent to an extract
subvector where the inner vector is any-extended to the
extract_vector_elt VT, because extract_vector_elt has the effect of an
any-extend.
(build_vector (extract_elt_i16_to_i32 vec Idx+0) (extract_elt_i16_to_i32 vec Idx+1))
=> (extract_subvector (anyext_i16_to_i32 vec) Idx)
Depends on D130697
Differential Revision: https://reviews.llvm.org/D130698
Without this, the intrinsic will be expanded to an integer; thereby an
explicit copy (from GPR to SIMD register) will be codegen'd. This matches the
general convention of using "v1" types to represent scalar integer operations in
vector registers.
The similar approach is observed in D56616, and the pattern likely applies on
other intrinsic that accepts integer scalars (e.g.,
int_aarch64_neon_sqdmulls_scalar)
Differential Revision: https://reviews.llvm.org/D130548
This adds similar heuristics to G_GLOBAL_VALUE, querying the cost of
materializing a specific constant in code size. Doing so prevents us from
sinking constants which require multiple instructions to generate into
use blocks.
Code size savings on CTMark -Os:
Program size.__text
before after diff
ClamAV/clamscan 381940.00 382052.00 0.0%
lencod/lencod 428408.00 428428.00 0.0%
SPASS/SPASS 411868.00 411876.00 0.0%
kimwitu++/kc 449944.00 449944.00 0.0%
Bullet/bullet 463588.00 463556.00 -0.0%
sqlite3/sqlite3 284696.00 284668.00 -0.0%
consumer-typeset/consumer-typeset 414492.00 414424.00 -0.0%
7zip/7zip-benchmark 595244.00 594972.00 -0.0%
mafft/pairlocalalign 247512.00 247368.00 -0.1%
tramp3d-v4/tramp3d-v4 372884.00 372044.00 -0.2%
Geomean difference -0.0%
Differential Revision: https://reviews.llvm.org/D130554
This patch starts small, only detecting sequences of the form
<a, a+n, a+2n, a+3n, ...> where a and n are ConstantSDNodes.
Differential Revision: https://reviews.llvm.org/D125194
This helps fold away the ptest instructions, which needs the knowledge on whether
the general predicate is known to zero the inactive lanes.
This fixes some PTEST regressions introduced by D129282.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D129852
Currently, when llvm-objdump is disassembling a code section and
encounters a point where no instruction can be decoded, it uses the
same policy on all targets: consume one byte of the section, emit it
as "<unknown>", and try disassembling from the next byte position.
On an architecture where instructions are always 4 bytes long and
4-byte aligned, this makes no sense at all. If a 4-byte word cannot be
decoded as an instruction, then the next place that a valid
instruction could //possibly// be found is 4 bytes further on.
Disassembling from a misaligned address can't possibly produce
anything that the code generator intended, or that the CPU would even
attempt to execute.
This patch introduces a new MCDisassembler virtual method called
`suggestBytesToSkip`, which allows each target to choose its own
resynchronization policy. For Arm (as opposed to Thumb) and AArch64,
I've filled in the new method to return a fixed width of 4.
Thumb is a more interesting case, because the criterion for
identifying 2-byte and 4-byte instruction encodings is very simple,
and doesn't require the particular instruction to be recognized. So
`suggestBytesToSkip` is also passed an ArrayRef of the bytes in
question, so that it can take that into account. The new test case
shows Thumb disassembly skipping over two unrecognized instructions,
and identifying one as 2-byte and one as 4-byte.
For targets other than Arm and AArch64, this is NFC: the base class
implementation of `suggestBytesToSkip` still returns 1, so that the
existing behavior is unchanged. Other targets can fill in their own
implementations as they see fit; I haven't attempted to choose a new
behavior for each one myself.
I've updated all the call sites of `MCDisassembler::getInstruction` in
llvm-objdump, and also one in sancov, which was the only other place I
spotted the same idiom of `if (Size == 0) Size = 1` after a call to
`getInstruction`.
Reviewed By: DavidSpickett
Differential Revision: https://reviews.llvm.org/D130357
Due to the way fixed length SVE lowering works, we sometimes introduce
ext/trunc nodes very late, these nodes then immediately get converted
into target specific nodes (UUNPKLO/UZP1) before they get a chance to be
folded into a load/store.
This patch introduces target specific dag combines for these nodes so that
we can still create extending loads/truncating stores out of them.
Differential Revision: https://reviews.llvm.org/D128065
This patch recognizes f16 immediates as legal and adds the necessary
patterns. This allows the fadda folding introduced in 05d424d165
to be applied to the f16 cases.
Differential Revision: https://reviews.llvm.org/D129989
When lowering add(a, select(mask, b, splat(0))) the sel instruction can
be removed by using predicated add/sub instructions.
Reviewed By: paulwalker-arm
Differential Revision: https://reviews.llvm.org/D129751
An async suspend models the split between two partial async functions.
`llvm.swift.async.context.addr ` will have a different value in the two
partial functions so it is not correct to generally CSE the instruction.
rdar://97336162
Differential Revision: https://reviews.llvm.org/D130201
As noticed on D129765 and reported on Issue #56531 - aarch64 targets can use the neon ctpop + add-reduce instructions to speed up scalar ctpop instructions, but we fail to do this for parity calculations.
I'm not sure where the cutoff should be for specific CPUs, but i64 (+ i128 special case) shows a definite reduction in instruction count. i32 is about the same (but scalar <-> neon transfers are probably more costly?), and sub-i32 promotion looks to be a definite regression compared to parity expansion optimized for those widths.
Differential Revision: https://reviews.llvm.org/D130246
GEP's across basic blocks were not getting splitted due to EnableGEPOpt
which was turned off by default. Hence, EarlyCSE missed the opportunity
to eliminate common part of GEP's. This can be achieved by simply
turning GEP pass on.
- This patch moves SeparateConstOffsetFromGEPPass() just before LSR.
- It enables EnableGEPOpt by default.
Resolves - https://github.com/llvm/llvm-project/issues/50528
Added an unit test.
Differential Revision: https://reviews.llvm.org/D128582
OverMighty