Lower a = b * C where C = (2^n + 1) * 2^m to
add w0, w0, w0, lsl n
lsl w0, w0, m
Differential Revision: https://reviews.llvm.org/D229245
llvm-svn: 287019
The wave barrier represents the discardable barrier. Its main purpose is to
carry convergent attribute, thus preventing illegal CFG optimizations. All lanes
in a wave come to convergence point simultaneously with SIMT, thus no special
instruction is needed in the ISA. The barrier is discarded during code generation.
Differential Revision: https://reviews.llvm.org/D26585
llvm-svn: 287007
Also, fix the test params to use an attribute rather than a CPU model
and remove the AVX run because that does nothing but check for a 'v'
prefix in all of these tests.
llvm-svn: 287003
In RateRegister of existing LSR, if a formula contains a Reg which is a SCEVAddRecExpr,
and this SCEVAddRecExpr's loop is an outerloop, the formula will be marked as Loser
and dropped.
Suppose we have an IR that %for.body is outerloop and %for.body2 is innerloop. LSR only
handle inner loop now so only %for.body2 will be handled.
Using the logic above, formula like
reg(%array) + reg({1,+, %size}<%for.body>) + 1*reg({0,+,1}<%for.body2>) will be dropped
no matter what because reg({1,+, %size}<%for.body>) is a SCEVAddRecExpr type reg related
with outerloop. Only formula like
reg(%array) + 1*reg({{1,+, %size}<%for.body>,+,1}<nuw><nsw><%for.body2>) will be kept
because the SCEVAddRecExpr related with outerloop is folded into the initial value of the
SCEVAddRecExpr related with current loop.
But in some cases, we do need to share the basic induction variable
reg{0 ,+, 1}<%for.body2> among LSR Uses to reduce the final total number of induction
variables used by LSR, so we don't want to drop the formula like
reg(%array) + reg({1,+, %size}<%for.body>) + 1*reg({0,+,1}<%for.body2>) unconditionally.
From the existing comment, it tries to avoid considering multiple level loops at the same time.
However, existing LSR only handles innermost loop, so for any SCEVAddRecExpr with a loop other
than current loop, it is an invariant and will be simple to handle, and the formula doesn't have
to be dropped.
Differential Revision: https://reviews.llvm.org/D26429
llvm-svn: 286999
Summary:
This fixes the runtime results produces by the fallback multiplication expansion introduced in r270720.
For tests I created a fuzz tester that compares the results with Boost.Multiprecision.
Reviewers: hfinkel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26628
llvm-svn: 286998
This patch helps avoids poor legalization of boolean vector results (e.g. 8f32 -> 8i1 -> 8i16) that feed into SINT_TO_FP by inserting an early SIGN_EXTEND and so help improve the truncation logic.
This is not necessary for AVX512 targets where boolean vectors are legal - AVX512 manages to lower ( sint_to_fp vXi1 ) into some form of ( select mask, 1.0f , 0.0f ) in most cases.
Fix for PR13248
Differential Revision: https://reviews.llvm.org/D26583
llvm-svn: 286979
This patch implements all the overloads for vec_xl_be and vec_xst_be. On BE,
they behaves exactly the same with vec_xl and vec_xst, therefore they are
simply implemented by defining a matching macro. On LE, they are implemented
by defining new builtins and intrinsics. For int/float/long long/double, it
is just a load (lxvw4x/lxvd2x) or store(stxvw4x/stxvd2x). For char/char/short,
we also need some extra shuffling before or after call the builtins to get the
desired BE order. For int128, simply call vec_xl or vec_xst.
llvm-svn: 286967
Summary:
Fix a case where the overflow value of type i1, which is legal on AVX512, was assigned to a VK1 register class.
We always want this value to be assigned to a GPR since the overflow return value is lowered to a SETO instruction.
Fixes pr30981.
Reviewers: mkuper, igorb, craig.topper, guyblank, qcolombet
Subscribers: qcolombet, llvm-commits
Differential Revision: https://reviews.llvm.org/D26620
llvm-svn: 286958
This patch adds the Sched Machine Model for Cortex-R52.
Details of the pipeline and descriptions are in comments
in file ARMScheduleR52.td included in this patch.
Reviewers: rengolin, jmolloy
Differential Revision: https://reviews.llvm.org/D26500
llvm-svn: 286949
Summary:
Add basic functionality to support call lowering for X86.
Currently only supports functions which return void and take zero arguments.
Inspired by commit 286573.
Reviewers: ab, qcolombet, t.p.northover
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26593
llvm-svn: 286935
Implement the Newton series for square root, its reciprocal and reciprocal
natively using the specialized instructions in AArch64 to perform each
series iteration.
Differential revision: https://reviews.llvm.org/D26518
llvm-svn: 286907
For example we were producing
push {r8, r10, r11, r4, r5, r7, lr}
This is misleading (r4, r5 and r7 are actually pushed before the rest), and
other components (stack folding recently) often forget to deal with the extra
complexity coming from the different order, leading to miscompiles. Finally, we
warn about our own code in -no-integrated-as mode without this, which is really
not a good idea.
Fixed usage of std::sort so that we (hopefully) use instantiations that
actually exist in GCC 4.8.
llvm-svn: 286881
Summary:
Replace a splat of zeros to a vector store by scalar stores of WZR/XZR.
The load store optimizer pass will merge them to store pair stores.
This should be better than a movi to create the vector zero followed by
a vector store if the zero constant is not re-used, since one
instructions and one register live range will be removed.
For example, the final generated code should be:
stp xzr, xzr, [x0]
instead of:
movi v0.2d, #0
str q0, [x0]
Reviewers: t.p.northover, mcrosier, MatzeB, jmolloy
Subscribers: aemerson, rengolin, llvm-commits
Differential Revision: https://reviews.llvm.org/D26561
llvm-svn: 286875
For example we were producing
push {r8, r10, r11, r4, r5, r7, lr}
This is misleading (r4, r5 and r7 are actually pushed before the rest), and
other components (stack folding recently) often forget to deal with the extra
complexity coming from the different order, leading to miscompiles. Finally, we
warn about our own code in -no-integrated-as mode without this, which is really
not a good idea.
llvm-svn: 286866
add an intrinsic to expose the 'VSX Scalar Convert Half-Precision to
Single-Precision' instruction.
Differential review: https://reviews.llvm.org/D26536
llvm-svn: 286862
Summary:
Extend image intrinsics to support data types of V1F32 and V2F32.
TODO: we should define a mapping table to change the opcode for data type of V2F32 but just one channel is active,
even though such case should be very rare.
Reviewers:
tstellarAMD
Differential Revision:
http://reviews.llvm.org/D26472
llvm-svn: 286860
The Stack slot coloring pass removes a store that is followed by a load
that deal with the same stack slot. The function isLoadFromStackSlot
is supposed to consider the loads that have no side-effects. This
patch fixed the issue by removing the unsafe loads from this function
Eg:
%vreg0<def> = L2_loadruh_io <fi#15>, 0
S2_storeri_io <fi#15>, 0, %vreg0
In this case, we load an unsigned extended half word and store this in to
the same stack slot. The Stack slot coloring pass considers safe to remove
the store. This patch marked all the non-vector byte and half word loads as
unsafe.
llvm-svn: 286843
-Don't print the 'x' suffix for the 128-bit reg/mem VEX encoded instructions in Intel syntax. This is consistent with the EVEX versions.
-Don't print the 'y' suffix for the 256-bit reg/reg VEX encoded instructions in Intel or AT&T syntax. This is consistent with the EVEX versions.
-Allow the 'x' and 'y' suffixes to be used for the reg/mem forms when we're assembling using Intel syntax.
-Allow the 'x' and 'y' suffixes on the reg/reg EVEX encoded instructions in Intel or AT&T syntax. This is consistent with what VEX was already allowing.
This should fix at least some of PR28850.
llvm-svn: 286787
nThis avoids the nasty problems caused by using
memory instructions that read the exec mask while
spilling / restoring registers used for control flow
masking, but only for VI when these were added.
This always uses the scalar stores when enabled currently,
but it may be better to still try to spill to a VGPR
and use this on the fallback memory path.
The cache also needs to be flushed before wave termination
if a scalar store is used.
llvm-svn: 286766
These will be used to replace the masked intrinsics so that InstCombineCalls can optimize the AVX-512 variable shifts the same way it does for AVX2.
llvm-svn: 286754
After this I'll add the unmasked intrinsics to InstCombineCalls to finish making our handling of these types of shuffles consistent between AVX-512 and the legacy intrinsics.
llvm-svn: 286725
Summary:
This is the first step towards being able to add the avx512 shift by immediate intrinsics to InstCombineCalls where we aleady support the sse2 and avx2 intrinsics. We need to the unmasked versions so we can avoid having to teach InstCombineCalls that it would need to insert selects sometimes. Instead we'll just add the selects around the new instrinsics in the frontend.
This change should also enable the shift by i32 intrinsics to take a non-constant shift value just like the avx2 and sse intrinsics. This will enable us to fix PR30691 once we update clang.
Next I'll switch clang to use the new builtins. Then we'll come back to the backend and remove/autoupgrade the old intrinsics. Then I'll work on the same series for variable shifts.
Reviewers: RKSimon, zvi, delena
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26333
llvm-svn: 286711
Summary: VALIGND and VALIGNQ are similar to PALIGNR but instead of working on a 128-bit lane they work on the entire vector register. This change leverages the shuffle rotate detection code used for PALIGNR to detect these cases.
Reviewers: delena, RKSimon
Subscribers: Farhana, llvm-commits
Differential Revision: https://reviews.llvm.org/D26297
llvm-svn: 286709
Summary:
This pass was assuming that when a PHI instruction defined a register
used by another PHI instruction that the defining insstruction would
be legalized before the using instruction.
This assumption was causing the pass to not legalize some PHI nodes
within divergent flow-control.
This fixes a bug that was uncovered by r285762.
Reviewers: nhaehnle, arsenm
Subscribers: kzhuravl, wdng, nhaehnle, yaxunl, tony-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D26303
llvm-svn: 286676
This patch corresponds to review:
https://reviews.llvm.org/D26480
Adds all the intrinsics used for various permute builtins that will
be added to altivec.h.
llvm-svn: 286638
Summary:
Fix off-by-one indexing error in loop checking that inserted value was a
splat vector.
Add code to check that INSERT_VECTOR_ELT nodes constructing the splat
vector have the expected constant index values.
Reviewers: t.p.northover, jmolloy, mcrosier
Subscribers: aemerson, llvm-commits, rengolin
Differential Revision: https://reviews.llvm.org/D26409
llvm-svn: 286616
This is a partial revert of r244615 (http://reviews.llvm.org/D11942),
which caused a major regression in debug info quality.
Turning the artificial __MergedGlobal symbols into private symbols
(l__MergedGlobal) means that the linker will not include them in the
symbol table of the final executable. Without a symbol table entry
dsymutil is not be able to process the debug info for any of the
merged globals and thus drops the debug info for all of them.
This patch is enabling the old behavior for all MachO targets while
leaving all other targets unaffected.
rdar://problem/29160481
https://reviews.llvm.org/D26531
llvm-svn: 286607
This patch corresponds to review:
https://reviews.llvm.org/D26307
Adds all the intrinsics used for various conversion builtins that will
be added to altivec.h. These are type conversions between various types of
vectors.
llvm-svn: 286596
This adds support for the compare logical and trap (memory)
instructions that were added as part of the miscellaneous
instruction extensions feature with zEC12.
llvm-svn: 286587
This adds support for the LZRF/LZRG/LLZRGF instructions that were
added on z13, and uses them for code generation were appropriate.
SystemZDAGToDAGISel::tryRISBGZero is updated again to prefer LLZRGF
over RISBG where both would be possible.
llvm-svn: 286586
This adds support for the 31-to-64-bit zero extension instructions
LLGT and LLGTR and uses them for code generation where appropriate.
Since this operation can also be performed via RISBG, we have to
update SystemZDAGToDAGISel::tryRISBGZero so that we prefer LLGT
over RISBG in case both are possible. The patch includes some
simplification to the tryRISBGZero code; this is not intended
to cause any (further) functional change in codegen.
llvm-svn: 286585
addSchedBarrierDeps() is supposed to add use operands to the ExitSU
node. The current implementation adds uses for calls/barrier instruction
and the MBB live-outs in all other cases. The use
operands of conditional jump instructions were missed.
Also added code to macrofusion to set the latencies between nodes to
zero to avoid problems with the fusing nodes lingering around in the
pending list now.
Differential Revision: https://reviews.llvm.org/D25140
llvm-svn: 286544
There is no need to track dependencies for constant physregs, as they
don't change their value no matter in what order you read/write to them.
Differential Revision: https://reviews.llvm.org/D26221
llvm-svn: 286526
When copying to/from a constant register interferences can be ignored.
Also update the documentation for isConstantPhysReg() to make it more
obvious that this transformation is valid.
Differential Revision: https://reviews.llvm.org/D26106
llvm-svn: 286503
Currently runtime metadata is emitted as an ELF section with name .AMDGPU.runtime_metadata.
However there is a standard way to convey vendor specific information about how to run an ELF binary, which is called vendor-specific note element (http://www.netbsd.org/docs/kernel/elf-notes.html).
This patch lets AMDGPU backend emits runtime metadata as a note element in .note section.
Differential Revision: https://reviews.llvm.org/D25781
llvm-svn: 286502
We were failing to extract a constant splat shift value if the shifted value was being masked.
The (shl (and (setcc) N01CV) N1CV) -> (and (setcc) N01CV<<N1CV) combine was unnecessarily preventing this.
llvm-svn: 286454
Summary: This adds all of the CodeGen tests which currently pass.
Reviewers: arsenm, kparzysz
Subscribers: japaric, wdng
Differential Revision: https://reviews.llvm.org/D26388
llvm-svn: 286418
For pairs of 32-bit registers: isub_lo, isub_hi.
For pairs of vector registers: vsub_lo, vsub_hi.
Add generic subreg indices: ps_sub_lo, ps_sub_hi, and a function
HexagonRegisterInfo::getHexagonSubRegIndex(RegClass, GenericSubreg)
that returns the appropriate subreg index for RegClass.
llvm-svn: 286377
This patch adds support for fptoui to 2i32 from both 2f64 and 2f32, building on Simon's change for the signed version in r284459 and using AVX-512 instructions.
If we don't have VLX support we need to use a 512-bit operation for v2f64->v2i32 and extract the result.
It also recognises that cvttpd2udq zeroes the upper 64-bits of the xmm result.
Differential Revision: https://reviews.llvm.org/D26331
llvm-svn: 286345
Summary: This allows the SSE intrinsic to use the EVEX instruction when available. It also fixes EVEX to not use a weird (v4i32 (fp_to_sint v2f64)) node and it merges some isel patterns. This also fixes some cases that weren't combining vzmovl with cvttpd2dq to remove extra moves.
Reviewers: delena, zvi, RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26330
llvm-svn: 286344
Summary:
This is needed to make the v64i8 and v32i16 types legal for the 512-bit VBMI instructions. Fixes PR30912.
Reviewers: delena, zvi
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26322
llvm-svn: 286339
The smallest tests that expose this are codegen tests (because SelectionDAGBuilder::visitSelect() uses matchSelectPattern
to create UMAX/UMIN nodes), but it's also possible to see the effects in IR alone with folds of min/max pairs.
If these were written as unsigned compares in IR, InstCombine canonicalizes the unsigned compares to signed compares.
Ie, running the optimizer pessimizes the codegen for this case without this patch:
define <4 x i32> @umax_vec(<4 x i32> %x) {
%cmp = icmp ugt <4 x i32> %x, <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
%sel = select <4 x i1> %cmp, <4 x i32> %x, <4 x i32> <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
ret <4 x i32> %sel
}
$ ./opt umax.ll -S | ./llc -o - -mattr=avx
vpmaxud LCPI0_0(%rip), %xmm0, %xmm0
$ ./opt -instcombine umax.ll -S | ./llc -o - -mattr=avx
vpxor %xmm1, %xmm1, %xmm1
vpcmpgtd %xmm0, %xmm1, %xmm1
vmovaps LCPI0_0(%rip), %xmm2 ## xmm2 = [2147483647,2147483647,2147483647,2147483647]
vblendvps %xmm1, %xmm0, %xmm2, %xmm0
Differential Revision: https://reviews.llvm.org/D26096
llvm-svn: 286318
Since IMPLIFIT_DEF instructions are omitted in the output, when the output
of an IMPLICIT_DEF instruction is stackified, the resulting register lacks
an explicit push, leading to a push/pop mismatch. Fix this by converting
such IMPLICIT_DEFs into CONST_I32 0 instructions so that they have explicit
pushes.
llvm-svn: 286274
Summary: In addition, the branch instructions will have proper BB destinations, not offsets, like before.
Reviewers: asl
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D23718
llvm-svn: 286252
Fixed an issue with vector usage of TargetLowering::isConstTrueVal / TargetLowering::isConstFalseVal boolean result matching.
The comment said we shouldn't handle constant splat vectors with undef elements. But the the actual code was returning false if the build vector contained no undef elements....
This patch now ignores the number of undefs (getConstantSplatNode will return null if the build vector is all undefs).
The change has also unearthed a couple of missed opportunities in AVX512 comparison code that will need to be addressed.
Differential Revision: https://reviews.llvm.org/D26031
llvm-svn: 286238
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
Under -enable-unsafe-fp-math, SELECT_CC lowering in AArch64
transforms floating point comparisons of the form "a == 0.0 ? 0.0 : x" to
"a == 0.0 ? a : x". But it incorrectly assumes that 'x' and 'a' have
the same type which can lead to a wrong CSEL node that crashes later
due to nonsensical copies.
Differential Revision: https://reviews.llvm.org/D26394
llvm-svn: 286231
Self-referencing PHI nodes need their destination operands to be constrained
because nothing else is likely to do so. For now we just pick a register class
naively.
Patch mostly by Ahmed again.
llvm-svn: 286183
Codegen prepare sinks comparisons close to a user is we have only one register
for conditions. For AMDGPU we have many SGPRs capable to hold vector conditions.
Changed BE to report we have many condition registers. That way IR LICM pass
would hoist an invariant comparison out of a loop and codegen prepare will not
sink it.
With that done a condition is calculated in one block and used in another.
Current behavior is to store workitem's condition in a VGPR using v_cndmask
and then restore it with yet another v_cmp instruction from that v_cndmask's
result. To mitigate the issue a forward propagation of a v_cmp 64 bit result
to an user is implemented. Additional side effect of this is that we may
consume less VGPRs in a cost of more SGPRs in case if holding of multiple
conditions is needed, and that is a clear win in most cases.
llvm-svn: 286171
Summary:
Some vector loads and stores generated from AArch64 intrinsics alias each other
unnecessarily, preventing better scheduling. We just need to transfer memory
operands during lowering.
Reviewers: mcrosier, t.p.northover, jmolloy
Subscribers: aemerson, rengolin, llvm-commits
Differential Revision: https://reviews.llvm.org/D26313
llvm-svn: 286168
Because we shift the stack pointer by an unknown amount, we need an
additional pointer. In the case where we have variable-size objects
as well, we can't reuse the frame pointer, thus three pointers.
Patch by Jacob Gravelle
Differential Revision: https://reviews.llvm.org/D26263
llvm-svn: 286160
If the branch was on a read-undef of vcc, passes that used
analyzeBranch to invert the branch condition wouldn't preserve
the undef flag resulting in a verifier error.
Fixes verifier failures in a future commit.
Also fix verifier error when inserting copy for vccz
corruption bug.
llvm-svn: 286133
When the base register (register pointing to the jump table) is the PC, we expect the jump table to directly follow the jump sequence with no intervening padding.
If there is intervening padding, the calculated offsets will not be correct. One solution would be to account for any padding in the emitted LDRB instruction, but at the moment we don't support emitting MCExprs for the load offset.
In the meantime, it's correct and only a slight amount worse to just move the padding up, from just before the jump table to just before the jump instruction sequence. We can do that by emitting code alignment before the jump sequence, as we know the number of instructions in the sequence is always 4.
llvm-svn: 286107
This handles the last case of the builtin function calls that we would
generate code which differed from Microsoft's ABI. Rather than
generating a call to `__pow{d,s}i2` we now promote the parameter to a
float or double and invoke `powf` or `pow` instead.
Addresses PR30825!
llvm-svn: 286082
Summary:
SmallSetVector uses DenseSet, but that means we need to reserve some
values for the empty and tombstone keys.
It seems to me we should have a general way to let us store full-range
ints inside of DenseSets, and furthermore that we probably shouldn't
silently let you add ints into DenseSets without explicitly promising
that they're in range. But that's a battle for another day; for now,
just fix this code, since we currently do something Very Bad when
compiling ffmpeg.
Fixes PR30914.
Reviewers: jeremyhu
Subscribers: llvm-commits, mzolotukhin
Differential Revision: https://reviews.llvm.org/D26323
llvm-svn: 286038
Summary: ARMv6m supports dmb etc fench instructions but not ldrex/strex etc. So for some atomic load/store, LLVM should inline instructions instead of lowering to __sync_ calls.
Reviewers: rengolin, efriedma, t.p.northover, jmolloy
Subscribers: efriedma, aemerson, llvm-commits
Differential Revision: https://reviews.llvm.org/D26120
llvm-svn: 285969
hange explores the fact that LDS reads may be reordered even if access
the same location.
Prior the change, algorithm immediately stops as soon as any memory
access encountered between loads that are expected to be merged
together. Although, Read-After-Read conflict cannot affect execution
correctness.
Improves hcBLAS CGEMM manually loop-unrolled kernels performance by 44%.
Also improvement expected on any massive sequences of reads from LDS.
Differential Revision: https://reviews.llvm.org/D25944
llvm-svn: 285919
This recommits r281323, which was backed out for two reasons. One, a selfhost failure, and two, it apparently caused Chromium failures. Actually, the latter was a red herring. The log has expired from the former, but I suspect that was a red herring too (actually caused by another problematic patch of mine). Therefore reapplying, and will watch the bots like a hawk.
For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)).
1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS.
2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS.
3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS).
4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask.
1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win.
llvm-svn: 285893
This fixes selection of KANDN instructions and allows us to remove an extra set of patterns for KNOT and KXNOR.
Reviewers: delena, igorb
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D26134
llvm-svn: 285878
2 new intrinsics covering AVX-512 compress/expand functionality.
This implementation includes syntax, DAG builder, operation lowering and tests.
Does not include: handling of illegal data types, codegen prepare pass and the cost model.
llvm-svn: 285876
Summary:
The post-RA scheduler occasionally uses additional implicit operands when
the vector implicit operand as a whole is killed, but some subregisters
are still live because they are directly referenced later. Unfortunately,
this seems incredibly subtle to reproduce.
Fixes piglit spec/glsl-110/execution/variable-indexing/vs-temp-array-mat2-index-wr.shader_test
and others.
Reviewers: arsenm, tstellarAMD
Subscribers: kzhuravl, wdng, yaxunl, tony-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D25656
llvm-svn: 285835
This is the conservatively correct way because it's easy to
move or replace a scalar immediate. This was incorrect in the case
when the register class wasn't known from the static instruction
definition, but still needed to be an SGPR. The main example of this
is inlineasm has an SGPR constraint.
Also start verifying the register classes of inlineasm operands.
llvm-svn: 285762
This will prevent following regression when enabling i16 support (D18049):
test/CodeGen/AMDGPU/ctlz.ll
test/CodeGen/AMDGPU/ctlz_zero_undef.ll
Differential Revision: https://reviews.llvm.org/D25802
llvm-svn: 285716
I wanted to implement this as a target independent expansion, however when
targets say they want to expand FP_TO_FP16 what they actually want is
the unsafe math expansion when possible and expansion to a libcall in all
other cases.
The only way to make this work as a target independent would be to add logic
to target's TargetLowering construction to mark theses nodes as Expand when
LegalizeDAG can use the unsafe expansion and mark them as LibCall when it
cannot. I think this would be possible, but I think it would be too fragile
and complex as it would require targets to keep their expansion logic up
to date with the code in LegalizeDAG.
Reviewers: bogner, ab, t.p.northover, arsenm
Subscribers: wdng, llvm-commits, nhaehnle
Differential Revision: https://reviews.llvm.org/D25999
llvm-svn: 285704
Note: Test is per differential review, but the other changed code in the review was for an optimisation that din't quite work. Nevertheless, the test is valid for the unoptimised version of the fix.
Differential Review: https://reviews.llvm.org/D24658
llvm-svn: 285692
[Reapplying r284580 and r285917 with fix and testing to ensure emitted jump tables for Thumb-1 have 4-byte alignment]
The TBB and TBH instructions in Thumb-2 allow jump tables to be compressed into sequences of bytes or shorts respectively. These instructions do not exist in Thumb-1, however it is possible to synthesize them out of a sequence of other instructions.
It turns out this sequence is so short that it's almost never a lose for performance and is ALWAYS a significant win for code size.
TBB example:
Before: lsls r0, r0, #2 After: add r0, pc
adr r1, .LJTI0_0 ldrb r0, [r0, #6]
ldr r0, [r0, r1] lsls r0, r0, #1
mov pc, r0 add pc, r0
=> No change in prologue code size or dynamic instruction count. Jump table shrunk by a factor of 4.
The only case that can increase dynamic instruction count is the TBH case:
Before: lsls r0, r4, #2 After: lsls r4, r4, #1
adr r1, .LJTI0_0 add r4, pc
ldr r0, [r0, r1] ldrh r4, [r4, #6]
mov pc, r0 lsls r4, r4, #1
add pc, r4
=> 1 more instruction in prologue. Jump table shrunk by a factor of 2.
So there is an argument that this should be disabled when optimizing for performance (and a TBH needs to be generated). I'm not so sure about that in practice, because on small cores with Thumb-1 performance is often tied to code size. But I'm willing to turn it off when optimizing for performance if people want (also note that TBHs are fairly rare in practice!)
llvm-svn: 285690
This bug was exposed by using nsw/nuw for more aggressive folds in:
https://reviews.llvm.org/rL284844
The changes mimic the IR demanded bits logic in InstCombiner::SimplifyDemandedUseBits(),
but we can't just flip flag bits in the DAG; we have to create a new node that has the
bits cleared.
This should fix:
https://llvm.org/bugs/show_bug.cgi?id=30841
llvm-svn: 285656
Generate the slowest possible codepath for noopt CodeGen. Even trying to be
clever with the negated jump can cause out-of-range jumps. Use a wide branch
instead. Although the code is modelled simplistically, the later optimizations
would recombine the branching into `cbz` if possible. This re-enables the
previous optimization as well as hopefully gives us working code in all cases.
Addresses PR30356!
llvm-svn: 285649
Summary:
This has been replaced by the NVPTXInferAddressSpaces pass. We've had
the new one as the default with the old one accessible via a flag for
some months now, and we've had no problems.
Reviewers: tra
Subscribers: llvm-commits, jholewinski, jingyue, mgorny
Differential Revision: https://reviews.llvm.org/D26165
llvm-svn: 285642
Try harder to detect obfuscated min/max patterns: the initial pattern was added with D9352 / rL236202.
There was a bug fix for PR27137 at rL264996, but I think we can do better by folding the corresponding
smax pattern and commuted variants.
The codegen tests demonstrate the effect of ValueTracking on the backend via SelectionDAGBuilder. We
can't expose these differences minimally in IR because we don't have smin/smax intrinsics for IR.
Differential Revision: https://reviews.llvm.org/D26091
llvm-svn: 285499
Currently computeKnownBits returns the common known zero/one bits for all elements of vector data, when we may only be interested in one/some of the elements.
This patch adds a DemandedElts argument that allows us to specify the elements we actually care about. The original computeKnownBits implementation calls with a DemandedElts demanding all elements to match current behaviour. Scalar types set this to 1.
The approach was found to be easier than trying to add a per-element known bits solution, for a similar usefulness given the combines where computeKnownBits is typically used.
I've only added support for a few opcodes so far (the ones that have proven straightforward to test), all others will default to demanding all elements but can be updated in due course.
DemandedElts support could similarly be added to computeKnownBitsForTargetNode in a future commit.
This looked like this had caused compile time regressions on some buildbots (and was reverted in rL285381), but appears to have just been a harmless bystander!
Differential Revision: https://reviews.llvm.org/D25691
llvm-svn: 285494
Instead of asserting that the shift count is != 0 we just bail out
as it's not profitable trying to optimize a node which will be
removed anyway.
Differential Revision: https://reviews.llvm.org/D26098
llvm-svn: 285480
Summary:
Flat instruction can return out of order, so we need always need to wait
for all the outstanding flat operations.
Reviewers: tony-tye, arsenm
Subscribers: kzhuravl, wdng, nhaehnle, llvm-commits, yaxunl
Differential Revision: https://reviews.llvm.org/D25998
llvm-svn: 285479
Also add glc bit to the scalar loads since they exist on VI
and change the caching behavior.
This currently has an assembler bug where the glc bit is incorrectly
accepted on SI/CI which do not have it.
llvm-svn: 285463
Summary:
In isel, transform
Num % Den
into
Num - (Num / Den) * Den
if the result of Num / Den is already available.
Reviewers: tra
Subscribers: hfinkel, llvm-commits, jholewinski
Differential Revision: https://reviews.llvm.org/D26090
llvm-svn: 285461
When LivePhysRegs adds live-in registers, it recognizes ~0 as a special
lane mask indicating the entire register. If the lane mask is not ~0,
it will only add the subregisters that overlap the specified lane mask.
The problem is that if a live-in register does not have subregisters,
and the lane mask is not ~0, it will not be added to the live set.
(The given lane mask may simply be the lane mask of its register class.)
If a register does not have subregisters, add it to the live set if
the lane mask is non-zero.
Differential Revision: https://reviews.llvm.org/D26094
llvm-svn: 285440
It's possible to have a use of the private resource descriptor or
scratch wave offset registers even though there are no allocated
stack objects. This would result in continuing to use the maximum
number reserved registers. This could go over the number of SGPRs
available on VI, or violate the SGPR limit requested by
the function attributes.
llvm-svn: 285435
Matt Arsenault