Going from XForm Load to DSForm Load requires that the immediate be 4 byte
aligned.
If we are not aligned we must leave the load as LDX (XForm).
This bug is causing a compile-time failure in the benchmark h264ref.
Differential Revision: https://reviews.llvm.org/D51988
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@343525 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
When PR16508 was solved (in rL185363) a regression test was
added as test/CodeGen/PowerPC/2013-07-01-PHIElimBug.ll.
I discovered that the test case no longer reproduced the
scenario from PR16508. This problem could have been amended
by adding an extra RUN line with "-O1" (or possibly "-O0"),
but instead I added a mir-reproducer
test/CodeGen/PowerPC/2013-07-01-PHIElimBug.mir
to get a reproducer that is less sensitive to changes in
earlier passes (including O-level).
While being at it I also corrected a code comment in
PHIElimination::EliminatePHINodes that has been incorrect
since the related bugfix from rL185363.
Reviewers: MatzeB, hfinkel
Reviewed By: MatzeB
Subscribers: nemanjai, jsji, llvm-commits
Differential Revision: https://reviews.llvm.org/D52553
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@343416 91177308-0d34-0410-b5e6-96231b3b80d8
This patch adds a check to optimize conditional branch (BC and BCn) based on a constant set by CRSET or CRUNSET.
Other optimizers, such as block placement, may generate such code and hence
I do this at the very end of the optimization in pre-emit peephole pass.
A conditional branch based on a constant is eliminated or converted into unconditional branch.
Also CRSET/CRUNSET is eliminated if the condition code register is not used
by instruction other than the branch to be optimized.
Differential Revision: https://reviews.llvm.org/D52345
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@343100 91177308-0d34-0410-b5e6-96231b3b80d8
gcc uses operand modifier 'x' in inline asm for VSX registers.
Without this modifier, instructions which use VSX numbering for their
operands are printed as VMX registers. This patch adds support for the
operand modifier 'x'.
Differential Revision: https://reviews.llvm.org/D52244
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342882 91177308-0d34-0410-b5e6-96231b3b80d8
Building a vector out of multiple loads can be converted to a load of the vector type if the loads are consecutive.
But the special condition is that the element number is 1, such as <1 x i128>. So just early exit to fix the assert.
Patch By: wuzish (Zixuan Wu)
Differential Revision: https://reviews.llvm.org/D52072
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342611 91177308-0d34-0410-b5e6-96231b3b80d8
This is a follow-up to the previous patch that eliminated some of the rotates.
With this addition, we will also emit the record-form andis.
This patch increases the number of record-form rotates we eliminate by
more than 70%.
Differential revision: https://reviews.llvm.org/D44897
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342478 91177308-0d34-0410-b5e6-96231b3b80d8
Both ANDIo and ANDISo (and the 64-bit versions) are record-form instructions.
When optimizing compares, we handle the former in order to eliminate the compare
instruction but not the latter. This patch just adds the latter to the set of
instructions we optimize.
The reason these instructions need to be handled separately is that they are not
part of the RecFormRel map (since they don't have a non-record-form). The
missing "and-immediate-shifted" is just an oversight in the initial
implementation.
Differential revision: https://reviews.llvm.org/D51353
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342472 91177308-0d34-0410-b5e6-96231b3b80d8
When doing some instruction scheduling work, we noticed some missing itineraries.
Before we switch to machine scheduler, those missing itineraries might not have impact to actually scheduling,
because we can still get same latency due to default values.
With machine scheduler, however, itineraries will have impact to scheduling.
eg: NumMicroOps will default to be 0 if there is NO itineraries for specific instruction class.
And most of the instruction class with itineraries will have NumMicroOps default to 1.
This will has impact on the count of RetiredMOps, affects the Pending/Available Queue,
then causing different scheduling or suboptimal scheduling further.
Patch By: jsji (Jinsong Ji)
Differential Revision: https://reviews.llvm.org/D52040
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342441 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
Integer types smaller than i32 must be extended to i32 by default.
The feature "crbits" introduced at r202451 handles i1 as a special case,
but it did not extend properly.
The caller was, therefore, passing i1 stack arguments by writing 0/1 to
the first byte of the 4-byte stack object and callee was
reading the first byte for the value.
"crbits" is enabled if the optimization level is greater than 1,
which is very common in "release builds".
Such discrepancies with ABI specification also introduces
potential incompatibility with programs or libraries
built with other compilers e.g. GCC.
Fixes PR38661
Reviewers: hfinkel, cuviper
Subscribers: sylvestre.ledru, glaubitz, nagisa, nemanjai, kbarton, llvm-commits
Differential Revision: https://reviews.llvm.org/D51108
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342288 91177308-0d34-0410-b5e6-96231b3b80d8
When doing some instruction scheduling work, we noticed some missing itineraries.
Before we switch to machine scheduler, those missing itineraries might not have impact to actually scheduling,
because we can still get same latency due to default values.
With machine scheduler, however, itineraries will have impact to scheduling.
eg: NumMicroOps will default to be 0 if there is NO itineraries for specific instruction class.
And most of the instruction class with itineraries will have NumMicroOps default to 1.
This will has impact on the count of RetiredMOps, affects the Pending/Available Queue,
then causing different scheduling or suboptimal scheduling further.
Patch by jsji (Jinsong Ji)
Differential Revision: https://reviews.llvm.org/D51506
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@341293 91177308-0d34-0410-b5e6-96231b3b80d8
This patch issues an error message if Darwin ABI is attempted with the PPC
backend. It also cleans up existing test cases, either converting the test to
use an alternative triple or removing the test if the coverage is no longer
needed.
Updated Tests
-------------
The majority of test cases were updated to use a different triple that does not
include the Darwin ABI. Many tests were also updated to use FileCheck, in place
of grep.
Deleted Tests
-------------
llvm/test/tools/dsymutil/PowerPC/sibling.test was originally added to test
specific functionality of dsymutil using an object file created with an old
version of llvm-gcc for a Powerbook G4. After a discussion with @JDevlieghere he
suggested removing the test.
llvm/test/CodeGen/PowerPC/combine_loads_from_build_pair.ll was converted from a
PPC test to a SystemZ test, as the behavior is also reproducible there.
All other tests that were deleted were specific to the darwin/ppc ABI and no
longer necessary.
Phabricator Review: https://reviews.llvm.org/D50988
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340795 91177308-0d34-0410-b5e6-96231b3b80d8
This commit has caused failures in some internal benchmarks. Temporarily
reverting this patch until the issue can be diagnosed and fixed.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340740 91177308-0d34-0410-b5e6-96231b3b80d8
The internal benchmark failure reported by Google was due to a missing
check for the result type for the sign-extend and shift DAG. This commit
adds the check and re-commits the patch.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340734 91177308-0d34-0410-b5e6-96231b3b80d8
This patch will address using the xscpsgndp instruction to copy floating point
scalar registers instead of the xxlor (specifically XXLORf) instruction that is
currently used. Additionally, this patch of utilizing xscpsgndp will apply to
P9, while pre-P9 will still use xxlor.
Patch by amyk
Differential Revision: https://reviews.llvm.org/D50004
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340643 91177308-0d34-0410-b5e6-96231b3b80d8
If the arch is P8, we will select XFLOAD to load the floating point, and then, expand it to vsx and non-vsx X-form instruction post RA. This patch is trying to convert the X-form to D-form if it meets the requirement that one operand of the x-form inst is the special Zero register, and another operand fed by add inst. i.e.
y = add imm, reg
LFDX. 0, y
-->
LFD imm(reg)
Reviewers: Nemanjai
Differential Revision: https://reviews.llvm.org/D49007
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340149 91177308-0d34-0410-b5e6-96231b3b80d8
This patch addresses:
- Implementation within PPCISelLowering.cpp to check if we should use direct
load into vector instructions (such as lxsd/lfd ) when the scalar_to_vector
function is used; which will allow us to catch as many cases of the
scalar_to_vector uses as possible to translate the ld->mtvsrd sequence into
lxsd.
- Test cases to exhibit the behaviour of emitting lxsd/lfd.
Patch by amyk
Differential revision: https://reviews.llvm.org/D49698
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@340037 91177308-0d34-0410-b5e6-96231b3b80d8
There is no way in the universe, that doing a full-width division in
software will be faster than doing overflowing multiplication in
software in the first place, especially given that this same full-width
multiplication needs to be done anyway.
This patch replaces the previous implementation with a direct lowering
into an overflowing multiplication algorithm based on half-width
operations.
Correctness of the algorithm was verified by exhaustively checking the
output of this algorithm for overflowing multiplication of 16 bit
integers against an obviously correct widening multiplication. Baring
any oversights introduced by porting the algorithm to DAG, confidence in
correctness of this algorithm is extremely high.
Following table shows the change in both t = runtime and s = space. The
change is expressed as a multiplier of original, so anything under 1 is
“better” and anything above 1 is worse.
+-------+-----------+-----------+-------------+-------------+
| Arch | u64*u64 t | u64*u64 s | u128*u128 t | u128*u128 s |
+-------+-----------+-----------+-------------+-------------+
| X64 | - | - | ~0.5 | ~0.64 |
| i686 | ~0.5 | ~0.6666 | ~0.05 | ~0.9 |
| armv7 | - | ~0.75 | - | ~1.4 |
+-------+-----------+-----------+-------------+-------------+
Performance numbers have been collected by running overflowing
multiplication in a loop under `perf` on two x86_64 (one Intel Haswell,
other AMD Ryzen) based machines. Size numbers have been collected by
looking at the size of function containing an overflowing multiply in
a loop.
All in all, it can be seen that both performance and size has improved
except in the case of armv7 where code size has regressed for 128-bit
multiply. u128*u128 overflowing multiply on 32-bit platforms seem to
benefit from this change a lot, taking only 5% of the time compared to
original algorithm to calculate the same thing.
The final benefit of this change is that LLVM is now capable of lowering
the overflowing unsigned multiply for integers of any bit-width as long
as the target is capable of lowering regular multiplication for the same
bit-width. Previously, 128-bit overflowing multiply was the widest
possible.
Patch by Simonas Kazlauskas!
Differential Revision: https://reviews.llvm.org/D50310
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@339922 91177308-0d34-0410-b5e6-96231b3b80d8
To make ISD::VSELECT available(legal) so long as there are altivec instruction,
otherwise it's default behavior is expanding.
Use xxsel to match vselect if vsx is open, or use vsel.
In order to do not write many patterns in td file, promote (for vector it's
bitcast) all other type into v4i32 and only pattern match vselect of v4i32 into
vsel or xxsel.
Patch by wuzish
Differential revision: https://reviews.llvm.org/D49531
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@339779 91177308-0d34-0410-b5e6-96231b3b80d8
When trying to combine a DAG that builds a vector out of sign-extensions of
vector extracts, the code assumes legal input types. Due to that, we have to
disable this combine prior to legalization.
In some cases, the DAG will look slightly different after legalization so
account for that in the matching code.
This is a fix for https://bugs.llvm.org/show_bug.cgi?id=38087
Differential Revision: https://reviews.llvm.org/D49080
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@339769 91177308-0d34-0410-b5e6-96231b3b80d8
Similar to rL337966 - if the DAGCombiner's rotate matching was
working as expected, I don't think we'd see any test diffs here.
AArch only goes right, and PPC only goes left.
x86 has both, so no diffs there.
Differential Revision: https://reviews.llvm.org/D50091
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@339359 91177308-0d34-0410-b5e6-96231b3b80d8
This patch aims to improve the codegen for vector loads involving the
scalar_to_vector (load X) sequence. Initially, ld->mv instructions were used
for scalar_to_vector (load X), so this patch allows scalar_to_vector (load X)
to utilize:
LXSD and LXSDX for i64 and f64
LXSIWAX for i32 (sign extension to i64)
LXSIWZX for i32 and f64
Committing on behalf of Amy Kwan.
Differential Revision: https://reviews.llvm.org/D48950
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@339260 91177308-0d34-0410-b5e6-96231b3b80d8
Adding the FP_ROUND nodes when combining FP_TO_[SU]INT of elements
feeding a BUILD_VECTOR into an FP_TO_[SU]INT of the built vector
loses precision. This patch removes the code that adds these nodes
to true f64 operands. It also adds patterns required to ensure
the code is still vectorized rather than converting individual
elements and inserting into a vector.
Fixes https://bugs.llvm.org/show_bug.cgi?id=38342
Differential Revision: https://reviews.llvm.org/D50121
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338658 91177308-0d34-0410-b5e6-96231b3b80d8
The bug is visible in the constant-folded x86 tests. We can't use the
negated shift amount when the type is not power-of-2:
https://rise4fun.com/Alive/US1r
...so in that case, use the regular lowering that includes a select
to guard against a shift-by-bitwidth. This path is improved by only
calculating the modulo shift amount once now.
Also, improve the rotate (with power-of-2 size) lowering to use
a negate rather than subtract from bitwidth. This improves the
codegen whether we have a rotate instruction or not (although
we can still see that we're not matching to a legal rotate in
all cases).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338592 91177308-0d34-0410-b5e6-96231b3b80d8
This is exchanging a sub-of-1 with add-of-minus-1:
https://rise4fun.com/Alive/plKAH
This is another step towards improving select-of-constants codegen (see D48970).
x86 is the motivating target, and those diffs all appear to be wins. PPC and AArch64 look neutral.
I've limited this to early combining (!LegalOperations) in case a target wants to reverse it, but
I think canonicalizing to 'add' is more likely to produce further transforms because we have more
folds for 'add'.
Differential Revision: https://reviews.llvm.org/D49924
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338317 91177308-0d34-0410-b5e6-96231b3b80d8
The tests with a constant sub operand were added with rL338143,
but the potential transform doesn't have that requirement, so
adding more tests with variable operands.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338150 91177308-0d34-0410-b5e6-96231b3b80d8
This is a follow-up suggested in D48970.
Alive proofs:
https://rise4fun.com/Alive/sII
We can eliminate an instruction in the usual select-of-constants
to bit hack transform by adjusting the add/sub with constant.
This is always a win.
There are more transforms that are likely wins, but they may need
target hooks in case some targets do not benefit.
This is another step towards making up for canonicalizing to
select-of-constants in rL331486.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338132 91177308-0d34-0410-b5e6-96231b3b80d8
Summary:
The Signal Processing Engine (SPE) is found on NXP/Freescale e500v1,
e500v2, and several e200 cores. This adds support targeting the e500v2,
as this is more common than the e500v1, and is in SoCs still on the
market.
This patch is very intrusive because the SPE is binary incompatible with
the traditional FPU. After discussing with others, the cleanest
solution was to make both SPE and FPU features on top of a base PowerPC
subset, so all FPU instructions are now wrapped with HasFPU predicates.
Supported by this are:
* Code generation following the SPE ABI at the LLVM IR level (calling
conventions)
* Single- and Double-precision math at the level supported by the APU.
Still to do:
* Vector operations
* SPE intrinsics
As this changes the Callee-saved register list order, one test, which
tests the precise generated code, was updated to account for the new
register order.
Reviewed by: nemanjai
Differential Revision: https://reviews.llvm.org/D44830
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337347 91177308-0d34-0410-b5e6-96231b3b80d8
As discussed here:
http://lists.llvm.org/pipermail/llvm-dev/2018-May/123292.htmlhttp://lists.llvm.org/pipermail/llvm-dev/2018-July/124400.html
We want to add rotate intrinsics because the IR expansion of that pattern is 4+ instructions,
and we can lose pieces of the pattern before it gets to the backend. Generalizing the operation
by allowing 2 different input values (plus the 3rd shift/rotate amount) gives us a "funnel shift"
operation which may also be a single hardware instruction.
Initially, I thought we needed to define new DAG nodes for these ops, and I spent time working
on that (much larger patch), but then I concluded that we don't need it. At least as a first
step, we have all of the backend support necessary to match these ops...because it was required.
And shepherding these through the IR optimizer is the primary concern, so the IR intrinsics are
likely all that we'll ever need.
There was also a question about converting the intrinsics to the existing ROTL/ROTR DAG nodes
(along with improving the oversized shift documentation). Again, I don't think that's strictly
necessary (as the test results here prove). That can be an efficiency improvement as a small
follow-up patch.
So all we're left with is documentation, definition of the IR intrinsics, and DAG builder support.
Differential Revision: https://reviews.llvm.org/D49242
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337221 91177308-0d34-0410-b5e6-96231b3b80d8
This is almost the same as an existing IR canonicalization in instcombine,
so I'm assuming this is a good early generic DAG combine too.
The motivation comes from reduced bit-hacking for select-of-constants in IR
after rL331486. We want to restore that functionality in the DAG as noted in
the commit comments for that change and the llvm-dev discussion here:
http://lists.llvm.org/pipermail/llvm-dev/2018-July/124433.html
The PPC and AArch tests show that those targets are already doing something
similar. x86 will be neutral in the minimal case and generally better when
this pattern is extended with other ops as shown in the signbit-shift.ll tests.
Note the asymmetry: we don't include the (extend (ifneg X)) transform because
it already exists in SimplifySelectCC(), and that is verified in the later
unchanged tests in the signbit-shift.ll files. Without the 'not' op, the
general transform to use a shift is always a win because that's a single
instruction.
Alive proofs:
https://rise4fun.com/Alive/ysli
Name: if pos, get -1
%c = icmp sgt i16 %x, -1
%r = sext i1 %c to i16
=>
%n = xor i16 %x, -1
%r = ashr i16 %n, 15
Name: if pos, get 1
%c = icmp sgt i16 %x, -1
%r = zext i1 %c to i16
=>
%n = xor i16 %x, -1
%r = lshr i16 %n, 15
Differential Revision: https://reviews.llvm.org/D48970
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337130 91177308-0d34-0410-b5e6-96231b3b80d8
Revision r322373 fixed a bug in how we materialize constants when the CR-field
needs to be set.
However the fix is overly conservative. It will only do the transform if
AND-ing the input with the new constant produces the same new constant.
This is of course correct, but not necessarily required.
If there are no futher uses of the constant, the constant can be changed.
If there are no uses of the GPR result, the final result of the materialization
isn't important other than it needs to compare to zero correctly (lt, gt, eq).
Differential revision: https://reviews.llvm.org/D42109
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@337008 91177308-0d34-0410-b5e6-96231b3b80d8
Add the two options -ppc-vsr-nums-as-vr and -ppc-asm-full-reg-names to
the __float128 tests. Then modify the tests as required.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@336940 91177308-0d34-0410-b5e6-96231b3b80d8
See https://reviews.llvm.org/D47106 for details.
Reviewed By: probinson
Differential Revision: https://reviews.llvm.org/D47171
This commit drops that patch's changes to:
llvm/test/CodeGen/NVPTX/f16x2-instructions.ll
llvm/test/CodeGen/NVPTX/param-load-store.ll
For some reason, the dos line endings there prevent me from commiting
via the monorepo. A follow-up commit (not via the monorepo) will
finish the patch.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@336843 91177308-0d34-0410-b5e6-96231b3b80d8
Stefan Pintilie