Summary: This patch originated from D46562 and is a proper subset, with some issues addressed for fsub.
Reviewers: spatel, hfinkel, wristow, arsenm
Reviewed By: spatel
Subscribers: wdng
Differential Revision: https://reviews.llvm.org/D47910
llvm-svn: 334306
As detailed on Agner's Microarchitecture doc (21.8 AMD Bobcat and Jaguar pipeline - Dependency-breaking instructions), these instructions are dependency breaking and fast-path zero the destination register (and appropriate EFLAGS bits).
llvm-svn: 334303
Same fix as rL334110: I noticed while working on zero-idiom + dependency-breaking support (PR36671) that most of our binary instruction schedule tests were reusing the same src registers, which would cause the tests to fail once we enable scalar zero-idiom support on btver2.
llvm-svn: 334302
AMDGPU inline assembler support i16, half and i128 typed variables in constraints, but they were reported as error.
Needed to fix https://github.com/RadeonOpenCompute/ROCm/issues/341,
e.g. to be able to load with global_load_dwordx4 to a 128bit integer variable
Differential Revision: https://reviews.llvm.org/D44920
llvm-svn: 334301
Summary:
`%ret = add nuw i8 %x, C`
From [[ https://llvm.org/docs/LangRef.html#add-instruction | langref ]]:
nuw and nsw stand for “No Unsigned Wrap” and “No Signed Wrap”,
respectively. If the nuw and/or nsw keywords are present,
the result value of the add is a poison value if unsigned
and/or signed overflow, respectively, occurs.
So if `C` is `-1`, `%x` can only be `0`, and the result is always `-1`.
I'm not sure we want to use `KnownBits`/`LVI` here, because there is
exactly one possible value (all bits set, `-1`), so some other pass
should take care of replacing the known-all-ones with constant `-1`.
The `test/Transforms/InstCombine/set-lowbits-mask-canonicalize.ll` change *is* confusing.
What happening is, before this: (omitting `nuw` for simplicity)
1. First, InstCombine D47428/rL334127 folds `shl i32 1, %NBits`) to `shl nuw i32 -1, %NBits`
2. Then, InstSimplify D47883/rL334222 folds `shl nuw i32 -1, %NBits` to `-1`,
3. `-1` is inverted to `0`.
But now:
1. *This* InstSimplify fold `%ret = add nuw i32 %setbit, -1` -> `-1` happens first,
before InstCombine D47428/rL334127 fold could happen.
Thus we now end up with the opposite constant,
and it is all good: https://rise4fun.com/Alive/OA9https://rise4fun.com/Alive/sldC
Was mentioned in D47428 review.
Follow-up for D47883.
Reviewers: spatel, craig.topper
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47908
llvm-svn: 334298
Simplify combineVectorTruncationWithPACKUS to mask the upper bits followed by calling truncateVectorWithPACK instead of duplicating with similar code.
This results in the codegen using (V)PACKUSDW on SSE41+ targets for vXi64/vXi32 inputs where before it always used PACKUSWB (along with a lot more bitcasting).
I've raised PR37749 as until we avoid unnecessary concats back to 256-bit for bitwise ops, we can't avoid splitting the input value into 128-bit subvectors for masking.
llvm-svn: 334289
Currently the loop branch heuristic is applied before the invoke heuristic which makes us overestimate the probability of the unwind destination of invokes inside loops. This in turn makes us grossly underestimate the frequencies of loops with invokes.
Reviewed By: skatkov, vsk
Differential Revision: https://reviews.llvm.org/D47371
llvm-svn: 334285
The instruction makes use of a previously ignored field in the fence
instruction. It is introduced in the version 2.3 draft of the RISC-V
specification after much work by the Memory Model Task Group.
As clarified here <https://github.com/riscv/riscv-isa-manual/issues/186>,
the fence.tso assembler mnemonic does not have operands.
llvm-svn: 334278
We have some combines/lowerings that attempt to use PACKSS-then-PACKUS and others that use PACKUS-then-PACKSS.
PACKUS is much easier to combine with if we know the upper bits are zero as ComputeKnownBits can easily see through BITCASTs etc. especially now that rL333995 and rL334007 have landed. It also effectively works at byte level which further simplifies shuffle combines.
The only (minor) annoyances are that ComputeKnownBits can sometimes take longer as it doesn't fail as quickly as ComputeNumSignBits (but I'm not seeing any actual regressions in tests) and PACKUSDW only became available after SSE41 so we have more codegen diffs between targets.
llvm-svn: 334276
There could be more than one PHIs in exit block using same loop recurrence.
Don't assume there is only one and fix each user.
Differential Revision: https://reviews.llvm.org/D47788
llvm-svn: 334271
While trying to propagate AND masks back to loads, we currently allow
one non-load node to be included as a leaf in chain. This fix now
limits that node to produce only a single data value.
Differential Revision: https://reviews.llvm.org/D47878
llvm-svn: 334268
Summary: This change uses fmf subflags to guard fma optimizations as well as unsafe. These changes originated from D46483 and have been simplified via getNode.
Reviewers: spatel, arsenm, hfinkel, javed.absar
Reviewed By: spatel
Subscribers: nemanjai, wdng
Differential Revision: https://reviews.llvm.org/D47388
llvm-svn: 334242
- Make code easier to maintain.
- Avoid generating waitcnts for VMEM if the address sppace does not involve VMEM.
- Add support to generate waitcnts for LDS and GDS memory.
Differential Revision: https://reviews.llvm.org/D47504
llvm-svn: 334241
%ret = add nuw i8 %x, C
From langref:
nuw and nsw stand for “No Unsigned Wrap” and “No Signed Wrap”,
respectively. If the nuw and/or nsw keywords are present,
the result value of the add is a poison value if unsigned
and/or signed overflow, respectively, occurs.
So if C is -1, %x can only be 0, and the result is always -1.
https://rise4fun.com/Alive/sldC
Was mentioned in D47428 review.
llvm-svn: 334236
Summary:
`%r = shl nuw i8 C, %x`
As per langref:
```
If the nuw keyword is present, then the shift produces
a poison value if it shifts out any non-zero bits.
```
Thus, if the sign bit is set on `C`, then `%x` can only be `0`,
which means that `%r` can only be `C`.
Or in other words, set sign bit means that the signed value
is negative, so the constant is `<= 0`.
https://rise4fun.com/Alive/WMkhttps://rise4fun.com/Alive/udv
Was mentioned in D47428 review.
We already handle the `0` constant, https://godbolt.org/g/UZq1sJ, so this only handles negative constants.
Could use computeKnownBits() / LazyValueInfo,
but the cost-benefit analysis (https://reviews.llvm.org/D47891)
suggests it isn't worth it.
Reviewers: spatel, craig.topper
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47883
llvm-svn: 334222
These weren't included in D19544 - probably just an oversight.
D40044 made it more likely that we'll have LLVM math intrinsics rather
than libcalls, so this bug was more easily exposed.
As the tests/code show, we already have the complete mappings for pow/exp/log.
I don't have any experience with SVML, so I don't know if anything else is
missing. It's also not clear to me that we should be doing this transform in
IR rather than DAG/isel, but that's a separate issue.
Differential Revision: https://reviews.llvm.org/D47610
llvm-svn: 334211
The implementation follows the MIPS backend and expands the pseudo instruction
directly during asm parsing. As the result, only real MC instructions are
emitted to the MCStreamer. The actual expansion to real instructions is
similar to the expansion performed by the GNU Assembler.
This patch supersedes D41949.
Differential Revision: https://reviews.llvm.org/D46118
Patch by Mario Werner.
llvm-svn: 334203
%r = shl nuw i8 C, %x
As per langref: If the nuw keyword is present, then the shift produces
a poison value if it shifts out any non-zero bits.
Thus, if the sign bit is set on C, then %x can only be 0,
which means that %r can only be C.
https://rise4fun.com/Alive/WMk
Was mentioned in D47428 review.
llvm-svn: 334200
BitPermutationSelector sets Repl32 flag for bit groups which can be (potentially) benefit from 32-bit rotate-and-mask instructions with bit replication, i.e. rlwinm/rlwimi copies lower 32 bits into upper 32 bits on 64-bit PowerPC before rotation.
However, enforcing 32-bit instruction sometimes results in redundant generated code.
For example, the following simple code is compiled into rotldi + rlwimi while it can be compiled into only rldimi instruction if Repl32 flag is not set on the bit group for (a & 0xFFFFFFFF).
uint64_t func(uint64_t a, uint64_t b) {
return (a & 0xFFFFFFFF) | (b << 32) ;
}
To avoid such problem, this patch checks the potential benefit of Repl32 flag before setting it. If a bit group does not require rotation (i.e. RLAmt == 0) and won't be merged into another group, we do not benefit from Repl32 flag on this group.
Differential Revision: https://reviews.llvm.org/D47867
llvm-svn: 334195
Simplify combineVectorTruncationWithPACKSS to just a SIGN_EXTEND_INREG followed by using the existing truncateVectorWithPACK instead of duplicating code.
llvm-svn: 334193
The existing trunc_shl_17_v8i16_v8i32 test case should (but doesn't) fold to zero, I've added 2 new test cases:
- trunc_shl_16_v8i16_v8i32 which folds to zero (this is actually testing the target faux shuffle combine)
- trunc_shl_15_v8i16_v8i32 which should perform the full shl + truncate
llvm-svn: 334188
We do the same thing in rewriteSingleStoreAlloca.
Fixes PR37632.
Reviewers: chandlerc, davide, efriedma
Reviewed By: davide
Differential Revision: https://reviews.llvm.org/D47825
llvm-svn: 334187
This has two main components. First, widen
widen short constant loads in DAG when they have
the correct alignment. This is already done a bit in
AMDGPUCodeGenPrepare, since that has access to
DivergenceAnalysis. This can't help kernarg loads
created in the DAG. Start to use DAG divergence analysis
to help this case.
The second part is to avoid kernel argument lowering
breaking the alignment of short vector elements because
calling convention lowering wants to split everything
into legal register types.
When loading a split type, load the nearest 4-byte aligned
segment and shift to get the desired bits. This extra
load of the earlier argument piece ends up merging,
and the bit extract hopefully folds out.
There are a number of improvements and regressions with
this, but I think as-is this is a better compromise between
several of the worst parts of SelectionDAG.
Particularly when i16 is legal, this produces worse code
for i8 and i16 element vector kernel arguments. This is
partially due to the very weak load merging the DAG does.
It only looks for fairly specific combines between pairs
of loads which no longer appear. In particular this
causes v4i16 loads to be split into 2 components when
previously the two halves were merged.
Worse, because of the newly introduced shifts, there
is a lot more unnecessary vector packing and unpacking code
emitted. At least some of this is due to reporting
false for isTypeDesirableForOp for i16 as a workaround for
the lack of divergence information in the DAG. The cases
where this happens it doesn't actually matter, but the
relevant code in SimplifyDemandedBits doens't have the context
to know to ignore this.
The use of the scalar cache is probably more important
than the mess of mostly scalar instructions doing this packing
and unpacking. Future work can fix this, possibly by making better
use of the new DAG divergence information for controlling promotion
decisions, or adding another version of shift + trunc + shift
combines that doesn't only know about the used types.
llvm-svn: 334180
Summary: Prevent folding of operations with memory loads when one of the sources has undefined register update.
Reviewers: craig.topper
Subscribers: llvm-commits, mike.dvoretsky, ashlykov
Differential Revision: https://reviews.llvm.org/D47621
llvm-svn: 334175
Summary:
When the branch folder hoist code into a predecessor it adjust live-in's
in the blocks it hoist code from. However it fail to handle hoisted code
that contain a defed register that originally is live-in in the block
through a super register.
This is fixed by replacing the live-in handling code with calls to
utility functions in LivePhysRegs.
Reviewers: kparzysz, gberry, MatzeB, uweigand, aprantl
Reviewed By: kparzysz
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47529
llvm-svn: 334163
With '-elf-output-style=GNU -relocations', a header containing the number
of entries is printed before all the relocation entries in the section.
For Android packed format, we need to perform the unpacking first before
we can get the actual number of relocations in the section.
Patch by Rahul Chaudhry!
Differential Revision: https://reviews.llvm.org/D47800
llvm-svn: 334147
When denormals are supported we are producing a full division for
1.0f / x. That still can be replaced by the faster version:
bool c = fabs(x) > 0x1.0p+96f;
float s = c ? 0x1.0p-32f : 1.0f;
x *= s;
return s * v_rcp_f32(x)
in case if requested accuracy is 2.5ulp or less. The same version
is used if denormals are not supported for non 1.0 numerators, where
just v_rcp_f32 is then used for 1.0 numerator.
The optimization of 1/x is extended to the case -1/x, which is the
same except for the resulting sign bit.
OpenCL conformance passed with both enabled and disabled denorms.
Differential Revision: https://reviews.llvm.org/D47805
llvm-svn: 334142
The bug report:
https://bugs.llvm.org/show_bug.cgi?id=36036
...requests a DAG change for this, but an IR canonicalization
probably handles most cases. If we still want to match this
pattern in the backend, there's a proposal for that too:
D47831
Alive proofs including nsw/nuw cases that were first noted in:
D46988
https://rise4fun.com/Alive/Kmp
This patch is largely copied from the existing code that was
initially added with:
D40984
...but I didn't see much gain from trying to share code.
llvm-svn: 334137
Fixes terrible code on targets without f16 support. The
legalization creates a mess that is difficult to recover
from. Also should avoid randomly breaking these tests
multiple times in sequence in future commits.
Some regressions in cases where it happens to be better
to pull the source modifier after the conversion.
llvm-svn: 334132
Summary:
This is [[ https://bugs.llvm.org/show_bug.cgi?id=37603 | PR37603 ]].
https://godbolt.org/g/VCMNpShttps://rise4fun.com/Alive/idM
When doing bit manipulations, it is quite common to calculate some bit mask,
and apply it to some value via `and`.
The typical C code looks like:
```
int mask_signed_add(int nbits) {
return (1 << nbits) - 1;
}
```
which is translated into (with `-O3`)
```
define dso_local i32 @mask_signed_add(int)(i32) local_unnamed_addr #0 {
%2 = shl i32 1, %0
%3 = add nsw i32 %2, -1
ret i32 %3
}
```
But there is a second, less readable variant:
```
int mask_signed_xor(int nbits) {
return ~(-(1 << nbits));
}
```
which is translated into (with `-O3`)
```
define dso_local i32 @mask_signed_xor(int)(i32) local_unnamed_addr #0 {
%2 = shl i32 -1, %0
%3 = xor i32 %2, -1
ret i32 %3
}
```
Since we created such a mask, it is quite likely that we will use it in `and` next.
And then we may get rid of `not` op by folding into `andn`.
But now that i have actually looked:
https://godbolt.org/g/VTUDmU
_some_ backend changes will be needed too.
We clearly loose `bzhi` recognition.
Reviewers: spatel, craig.topper, RKSimon
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D47428
llvm-svn: 334127
Michael Berg