We came across an llvm bug when compiling some testcases that 64-bit
immediates are silently truncated into 32-bit and then packed into
BPF_JMP | BPF_K encoding. This caused comparison with wrong value.
This bug looks to be introduced by r308080. The Select_Ri pattern is
supposed to be lowered into J*_Ri while the latter only support 32-bit
immediate encoding, therefore Select_Ri should have similar immediate
predicate check as what J*_Ri are doing.
Reported-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Reviewed-by: Yonghong Song <yhs@fb.com>
llvm-svn: 315889
This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass.
If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated.
One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI.
For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated.
void int_func(int);
void ii_test(int a) {
if (a & 1) return int_func(a);
}
Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG.
Differential Revision: https://reviews.llvm.org/D31319
llvm-svn: 315888
Summary:
iPTR is a pointer of subtarget-specific size to any address space. Therefore
type checks on this size derive the SizeInBits from a subtarget hook.
At this point, we can import the simplests G_LOAD rules and select load
instructions using them. Further patches will support for the predicates to
enable additional loads as well as the stores.
The previous commit failed on MSVC due to a failure to convert an
initializer_list to a std::vector. Hopefully, MSVC will accept this version.
Depends on D37457
Reviewers: ab, qcolombet, t.p.northover, rovka, aditya_nandakumar
Reviewed By: qcolombet
Subscribers: kristof.beyls, javed.absar, llvm-commits, igorb
Differential Revision: https://reviews.llvm.org/D37458
llvm-svn: 315887
Summary:
iPTR is a pointer of subtarget-specific size to any address space. Therefore
type checks on this size derive the SizeInBits from a subtarget hook.
At this point, we can import the simplests G_LOAD rules and select load
instructions using them. Further patches will support for the predicates to
enable additional loads as well as the stores.
Depends on D37457
Reviewers: ab, qcolombet, t.p.northover, rovka, aditya_nandakumar
Reviewed By: qcolombet
Subscribers: kristof.beyls, javed.absar, llvm-commits, igorb
Differential Revision: https://reviews.llvm.org/D37458
llvm-svn: 315885
Summary:
This includes some context-sensitivity in the MVT to LLT conversion so that
pointer types are tested correctly.
FIXME: I'm not happy with the way this is done since everything is a
special-case. I've yet to find a reasonable way to implement it.
select-load.mir fails because <1 x s64> loads in tablegen get priority over s64
loads. This is fixed in the next patch and as such they should be committed
together, I've posted them separately to help with the review.
Depends on D37456
Reviewers: ab, qcolombet, t.p.northover, rovka, aditya_nandakumar
Subscribers: kristof.beyls, javed.absar, llvm-commits, igorb
Differential Revision: https://reviews.llvm.org/D37457
llvm-svn: 315884
Summary:
This was impeding our ability to combine the extending shuffles with other shuffles as you can see from the test changes.
There's one special case that needed to be added to use VZEXT directly for v8i8->v8i64 since the custom lowering requires v64i8.
Reviewers: RKSimon, zvi, delena
Reviewed By: delena
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38714
llvm-svn: 315860
Summary:
It's better to use our shuffle lowering code to handle these than loading an immediate into a k-register.
It really feels like this should be a DAG combine optimization rather than a lowering operation, but that's a problem for another day.
Reviewers: RKSimon, delena, zvi
Reviewed By: delena
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38932
llvm-svn: 315849
We should be able to fold constant conditions by converting to shuffles, but fixing that would break these tests in their current form. Since they are really trying to test masking ops, add a non-constant mask to the selects.
llvm-svn: 315848
- Emit NT_AMD_AMDGPU_ISA
- Add assembler parsing for isa version directive
- If isa version directive does not match command line arguments, then return error
Differential Revision: https://reviews.llvm.org/D38748
llvm-svn: 315808
If we are applying a byte mask to a value extracted from a shuffle, see if we can combine the mask into shuffle.
Fixes the last issue with PR22415
llvm-svn: 315807
Summary:
Operand variable lookups are now performed by the RuleMatcher rather than
searching the whole matcher hierarchy for a match. This revealed a wrong-code
bug that currently affects ARM and X86 where patterns that use a variable more
than once in the match pattern will be imported but won't check that the
operands are identical. This can cause the tablegen-erated matcher to
accept matches that should be rejected.
Depends on D36569
Reviewers: ab, t.p.northover, qcolombet, rovka, aditya_nandakumar
Subscribers: aemerson, igorb, llvm-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D36618
llvm-svn: 315780
This is particularly important for AVX512VL where we are better able to recognize the VBROADCAST loads to fold with other operations.
For AVX512VL we now use X86ISD::VBROADCAST for all of the patterns and remove the 128-bit X86ISD::VMOVDDUP.
We may be able to use this for AVX1 as well which would allow us to remove more isel patterns.
I also had to add X86ISD::VBROADCAST as a node to call combineShuffle for so that we treat it similar to X86ISD::MOVDDUP.
Differential Revision: https://reviews.llvm.org/D38836
llvm-svn: 315768
Summary:
There's only a tablegen testcase for IntImmLeaf and not a CodeGen one
because the relevant rules are rejected for other reasons at the moment.
On AArch64, it's because there's an SDNodeXForm attached to the operand.
On X86, it's because the rule either emits multiple instructions or has
another predicate using PatFrag which cannot easily be supported at the
same time.
Reviewers: ab, t.p.northover, qcolombet, rovka, aditya_nandakumar
Reviewed By: qcolombet
Subscribers: aemerson, javed.absar, igorb, llvm-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D36569
llvm-svn: 315761
Each constant extender requires an extra instruction, which adds to the
code size and also reduces the number of available slots in an instruction
packet. In most cases, the value of a repeated constant extender could be
loaded into a register, and the instructions using the extender could be
replaced with their counterparts that use that register instead.
This patch adds a pass that tries to reduce the number of constant
extenders, including extenders which differ only in an immediate offset
known at compile time, e.g. @global and @global+12.
llvm-svn: 315735
This adds Intel's Knights Mill CPU to valid CPU names for the backend. For now its an alias of "knl", but ultimately we need to support AVX5124FMAPS and AVX5124VNNIW instruction sets for it.
Differential Revision: https://reviews.llvm.org/D38811
llvm-svn: 315722
There's no advantage to using these instructions when they aren't masked. This enables some additional execution domain switching without needing to update the table.
llvm-svn: 315674
WMMA = "Warp Level Matrix Multiply-Accumulate".
These are the new instructions introduced in PTX6.0 and available
on sm_70 GPUs.
Differential Revision: https://reviews.llvm.org/D38645
llvm-svn: 315601
Add profitability checks for modifying counted loops to use the mtctr instruction.
The latency of mtctr is only justified if there are more than 4 comparisons that
will be removed as a result. Usually counted loops are formed relatively early
and before unrolling, so most low trip count loops often don't survive. However
we want to ensure that if they do, we do not mistakenly update them to mtctr loops.
Use CodeMetrics to ensure we are only doing this for small loops with small trip counts.
Differential Revision: https://reviews.llvm.org/D38212
llvm-svn: 315592
Summary:
The interpolation mode workaround ensures that at least one
interpolation mode is enabled in PSInputAddr. It does not also check
PSInputEna on the basis that the user might enable bits in that
depending on run-time state.
However, for amdpal os type, the user does not enable some bits after
compilation based on run-time states; the register values being
generated here are the final ones set in the hardware. Therefore, apply
the workaround to PSInputAddr and PSInputEnable together. (The case
where a bit is set in PSInputAddr but not in PSInputEnable is where the
frontend set up an input arg for a particular interpolation mode, but
nothing uses that input arg. Really we should have an earlier pass that
removes such an arg.)
Reviewers: arsenm, nhaehnle, dstuttard
Subscribers: kzhuravl, wdng, yaxunl, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D37758
llvm-svn: 315591
Summary:
The comments in the code said
// Remove <def,read-undef> flags. This def is now a partial redef.
but the code didn't just remove read-undef, it could introduce new ones which
could cause errors.
E.g. if we have something like
%vreg1<def> = IMPLICIT_DEF
%vreg2:subreg1<def, read-undef> = op %vreg3, %vreg4
%vreg2:subreg2<def> = op %vreg6, %vreg7
and we merge %vreg1 and %vreg2 then we should not set undef on the second subreg
def, which the old code did.
Now we solve this by actually do what the code comment says. We remove
read-undef flags rather than remove or introduce them.
Reviewers: qcolombet, MatzeB
Reviewed By: MatzeB
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38616
llvm-svn: 315564
- Move PAL metadata definitions to AMDGPUMetadata
- Make naming consistent with HSA metadata
Differential Revision: https://reviews.llvm.org/D38745
llvm-svn: 315523
- Rename AMDGPUCodeObjectMetadata to AMDGPUMetadata (PAL metadata will be included in this file in the follow up change)
- Rename AMDGPUCodeObjectMetadataStreamer to AMDGPUHSAMetadataStreamer
- Introduce HSAMD namespace
- Other minor name changes in function and test names
llvm-svn: 315522
Summary:
Fixes a bogus iterator resulting from the removal of a block's first instruction at the point that incremental update is enabled.
Patch by Paul Walker.
Reviewers: fhahn, Gerolf, efriedma, MatzeB
Reviewed By: fhahn
Subscribers: aemerson, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D38734
llvm-svn: 315502
Currently we produce a bunch of unnecessary code when emitting the
prologue/epilogue for spills/restores. Namely, if the load from stack
slot/store to stack slot instruction is an X-Form instruction, we will
always produce an LIS/ORI sequence for the stack offset.
Furthermore, we have not exploited the P9 vector D-Form loads/stores for this
purpose.
This patch address both issues.
Specifying the D-Form load as the instruction to use for stack spills/reloads
should be safe because:
1. The stack should be aligned according to the ABI
2. If the stack isn't aligned, PPCRegisterInfo::eliminateFrameIndex() will
check for the offset being a multiple of 16 and will convert it to an
X-Form instruction if it isn't.
Differential Revision : https://reviews.llvm.org/D38758
llvm-svn: 315500
Legalization of fp128 assumes things that we should have asserts for,
so that's another potential improvement.
Differential Revision: https://reviews.llvm.org/D38771
llvm-svn: 315485
The pipeliner is generating a serial sequence that causes poor
register allocation when a post-increment instruction appears
prior to the use of the post-increment register. This occurs when
there is a circular set of dependences involved with a sequence
of instructions in the same cycle. In this case, there is no
serialization of the parallel semantics that will not cause an
additional register to be allocated.
This patch fixes the problem by changing the instructions so that
the post-increment instruction is used by the subsequent
instruction, which enables the register allocator to make a
better decision and not require another register.
Patch by Brendon Cahoon.
llvm-svn: 315466
Eg:
insert v4i32 V, (v2i16 X), 2 --> shuffle v8i16 V', X', {0,1,2,3,8,9,6,7}
This is a generalization of the IR fold in D38316 to handle insertion into a non-undef vector.
We may want to abandon that one if we can't find value in squashing the more specific pattern sooner.
We're using the existing legal shuffle target hook to avoid AVX512 horror with vXi1 shuffles.
There may be room for improvement in the shuffle lowering here, but that would be follow-up work.
Differential Revision: https://reviews.llvm.org/D38388
llvm-svn: 315460
Adding this test files now so after another commit that will add a new pattern for
TESTM and TESTNM instructions will show the improvemnts that have been done.
Change-Id: If3908b7f91897d764053312365a2bc1de78b291d
llvm-svn: 315443
Rather than using the AdditionalPredicates mechanism to guard
the microMIPS instructions, use the existing predicates to properly
guard those instructions.
This also resolves a case where an instruction pattern was incorrectly
available for microMIPS32R6, which caused a register allocation failure
as the registers specified in the pattern were not available.
Reviewers: nitesh.jain, atanasyan
Differential Revision: https://reviews.llvm.org/D38451
llvm-svn: 315362
opt-bisect/optnone disable the AMDGPUUniformAnnotateValues pass.
The heuristic in the custom selector for brcond deferred the
branch uniformity check to the pattern, which would fail.
llvm-svn: 315360
This patch adds a post-linking pass which replaces the function pointer of enqueued
block kernel with a global variable (runtime handle) and adds
runtime-handle attribute to the enqueued block kernel.
In LLVM CodeGen the runtime-handle metadata will be translated to
RuntimeHandle metadata in code object. Runtime allocates a global buffer
for each kernel with RuntimeHandel metadata and saves the kernel address
required for the AQL packet into the buffer. __enqueue_kernel function
in device library knows that the invoke function pointer in the block
literal is actually runtime handle and loads the kernel address from it
and puts it into AQL packet for dispatching.
This cannot be done in FE since FE cannot create a unique global variable
with external linkage across LLVM modules. The global variable with internal
linkage does not work since optimization passes will try to replace loads
of the global variable with its initialization value.
Differential Revision: https://reviews.llvm.org/D38610
llvm-svn: 315352
This patch ensures that the rule:
fold (zext (load x)) -> (zext (truncate (zextload x)))
propagates the SDLoc of the load to the zextload.
<rdar://problem/33755881>
llvm-svn: 315340
Summary:
The pass to fix function bitcasts generates thunks for functions that
are called directly with a mismatching signature. It was also generating
thunks in cases where the function was address-taken, causing aliasing
problems in otherwise valid cases.
This patch tightens the restrictions for when the pass runs.
Reviewers: sunfish, dschuff
Subscribers: jfb, sbc100, llvm-commits, aheejin
Differential Revision: https://reviews.llvm.org/D38640
llvm-svn: 315326
Summary:
See https://llvm.org/PR33743 for more details
It seems that for non-power of 2 vector sizes, the algorithm can produce
non-matching sizes for input and result causing an assert.
This usually isn't a problem as the isAnyExtend check will weed these out, but
in some cases (most often with lots of undefined values for the mask indices) it
can pass this check for non power of 2 vectors.
Adding in an extra check that ensures that bit size will match for the result
and input (as required)
Subscribers: nhaehnle
Differential Revision: https://reviews.llvm.org/D35241
llvm-svn: 315307
Summary:
Atomic buffer operations do not work (and trap on gfx9) when the
components are unaligned, even if their sum is aligned.
Previously, we generated an offset of 4156 without an SGPR by
splitting it as 4095 + 61 (immediate + inline constant). The
highest offset for which we can do this correctly is 4156 = 4092 + 64.
Fixes dEQP-GLES31.functional.ssbo.atomic.*
Reviewers: arsenm
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D37850
llvm-svn: 315302
NFC.
Updated 6 regression tests to differentiate between HASWELL and SKYLAKE scheduling information.
The fix is in preparation of a patch to update the information of the Skylake Client scheduling to include the appropriate load and store latencies.
Reviewers: zvi, RKSimon
Differential Revision: https://reviews.llvm.org/D38685
Change-Id: Ifc6b98d9eaf266913698f24c766fd994fc977555
llvm-svn: 315291
The issue is that we assume operand zero of the input to the add instruction
is a register. In this case, the input comes from inline assembly and
operand zero is not a register thereby causing a crash.
The code will bail anyway if the input instruction doesn't have the right
opcode. So do that check first and let short-circuiting prevent the crash.
llvm-svn: 315285
Summary:
This suppresses the generation of .Lcfi labels in our textual assembler.
It was annoying that this generated cascading .Lcfi labels:
llc foo.ll -o - | llvm-mc | llvm-mc
After three trips through MCAsmStreamer, we'd have three labels in the
output when none are necessary. We should only bother creating the
labels and frame data when making a real object file.
This supercedes D38605, which moved the entire .seh_ implementation into
MCObjectStreamer.
This has the advantage that we do more checking when emitting textual
assembly, as a minor efficiency cost. Outputting textual assembly is not
performance critical, so this shouldn't matter.
Reviewers: majnemer, MatzeB
Subscribers: qcolombet, nemanjai, javed.absar, eraman, hiraditya, JDevlieghere, llvm-commits
Differential Revision: https://reviews.llvm.org/D38638
llvm-svn: 315259
We end up creating COPY's that are either truncating/extending and this
should be illegal.
https://reviews.llvm.org/D37640
Patch for X86 and ARM by igorb, rovka
llvm-svn: 315240
Summary:
On behalf of julia.koval@intel.com
The patch transforms canonical version of unsigned saturation, which is sub(max(a,b),a) or sub(a,min(a,b)) to special psubus insturuction on targets, which support it(8bit and 16bit uints).
umax(a,b) - b -> subus(a,b)
a - umin(a,b) -> subus(a,b)
There is also extra case handled, when right part of sub is 32 bit and can be truncated, using UMIN(this transformation was discussed in https://reviews.llvm.org/D25987).
The example of special case code:
```
void foo(unsigned short *p, int max, int n) {
int i;
unsigned m;
for (i = 0; i < n; i++) {
m = *--p;
*p = (unsigned short)(m >= max ? m-max : 0);
}
}
```
Max in this example is truncated to max_short value, if it is greater than m, or just truncated to 16 bit, if it is not. It is vaid transformation, because if max > max_short, result of the expression will be zero.
Here is the table of types, I try to support, special case items are bold:
| Size | 128 | 256 | 512
| ----- | ----- | ----- | -----
| i8 | v16i8 | v32i8 | v64i8
| i16 | v8i16 | v16i16 | v32i16
| i32 | | **v8i32** | **v16i32**
| i64 | | | **v8i64**
Reviewers: zvi, spatel, DavidKreitzer, RKSimon
Reviewed By: zvi
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D37534
llvm-svn: 315237
E.g. if we have a (xor(overflow-bit), 1) where overflow-bit comes from an
intrinsic like llvm.sadd.with.overflow then we can kill the xor and use the
inverted condition code for the CSEL.
rdar://28495949
Reviewed By: kristof.beyls
Differential Revision: https://reviews.llvm.org/D38160
llvm-svn: 315205
We believe that despite AMD's documentation, that they really do support all 32 comparision predicates under AVX.
Differential Revision: https://reviews.llvm.org/D38609
llvm-svn: 315201
Summary:
We currently disable some converting of shuffles to MOVSS/MOVSD during legalization if SSE41 is enabled. But later during shuffle combining we go back to prefering MOVSS/MOVSD.
Additionally we have patterns that look for BLENDIs to detect scalar arithmetic operations. I believe due to the combining using MOVSS/MOVSD these are unnecessary.
Interestingly, we still codegen blend instructions even though lowering/isel emit movss/movsd instructions. Turns out machine CSE commutes them to blend, and then commuting those blends back into blends that are equivalent to the original movss/movsd.
This patch fixes the inconsistency in legalization to prefer MOVSS/MOVSD. The one test change was caused by this change. The problem is that we have integer types and are mostly selecting integer instructions except for the shufps. This shufps forced the execution domain, but the vpblendw couldn't have its domain changed with a naive instruction swap. We could fix this by special casing VPBLENDW based on the immediate to widen the element type.
The rest of the patch is removing all the excess scalar patterns.
Long term we should probably add isel patterns to make MOVSS/MOVSD emit blends directly instead of relying on the double commute. We may also want to consider emitting movss/movsd for optsize. I also wonder if we should still use the VEX encoded blendi instructions even with AVX512. Blends have better throughput, and that may outweigh the register constraint.
Reviewers: RKSimon, zvi
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38023
llvm-svn: 315181
Adding the scheduling information for the SkylakeServer (SKX) target.
This patch adds the instruction scheduling information for the SkylakeServer (SKX) architecture target by adding the file X86SchedSkylakeServer.td located under the X86 Target.
We used the scheduling information retrieved from the Skylake architects in order to create the file.
The scheduling information includes latency, number of micro-Ops and used ports by each SKL instruction.
The patch continues the scheduling replacement and insertion effort started with the SNB target in r310792, the HSW target in r311879 and the SkylakeClient (SKL) target in rL313613.
Please expect some performance fluctuations due to code alignment effects.
Reviewers: zvi, RKSimon, craig.topper, chandlerc, aymanmu
Differential Revision: https://reviews.llvm.org/D38443
Change-Id: I5c228fcc09e9e5a99b6116e62b356c4f9b971185
llvm-svn: 315175
Recognise cases when we can merge the shuffles with their horizontal (HADD/HSUB/PACK) instruction inputs.
Replaces an older implementation which performed some of this during lowering, expanding an existing target shuffle combine stage instead.
Differential Revision: https://reviews.llvm.org/D38506
llvm-svn: 315150
Say you have two identical linkonceodr functions, one in M1 and one in M2.
Say that the outliner outlines A,B,C from one function, and D,E,F from another
function (where letters are instructions). Now those functions are not
identical, and cannot be deduped. Locally to M1 and M2, these outlining
choices would be good-- to the whole program, however, this might not be true!
To mitigate this, this commit makes it so that the outliner sees linkonceodr
functions as unsafe to outline from. It also adds a flag,
-enable-linkonceodr-outlining, which allows the user to specify that they
want to outline from such functions when they know what they're doing.
Changing this handles most code size regressions in the test suite caused by
competing with linker dedupe. It also doesn't have a huge impact on the code
size improvements from the outliner. There are 6 tests that regress > 5% from
outlining WITH linkonceodrs to outlining WITHOUT linkonceodrs. Overall, most
tests either improve or are not impacted.
Not outlined vs outlined without linkonceodrs:
https://hastebin.com/raw/qeguxavuda
Not outlined vs outlined with linkonceodrs:
https://hastebin.com/raw/edepoqoqic
Outlined with linkonceodrs vs outlined without linkonceodrs:
https://hastebin.com/raw/awiqifiheb
Numbers generated using compare.py with -m size.__text. Tests run for AArch64
with -Oz -mllvm -enable-machine-outliner -mno-red-zone.
llvm-svn: 315136
Patch to fix ternlog instructions with a folded
broadcast. The broadcast decorator, e.g. {1toX}, was missing.
Differential Revision: https://reviews.llvm.org/D38649
llvm-svn: 315122
In some cases an instruction is deleted from the block during combining,
however it can still exist in the legalizer worklist.
This change modifies the combiner routines to add the given MI to the
dead instruction list rather than trying to remove it from the block
themselves. The responsibility is then on the caller to delete the instruction
from the block and any worklists.
Differential Revision: https://reviews.llvm.org/D38622
llvm-svn: 315092
Old expansion was 20 VGPRs, 78 SGPRs and ~380 instructions.
This expansion is 11 VGPRs, 12 SGPRs and ~120 instructions.
Passes OpenCL conformance test_integer_ops quick_[u]long_math
Differential Revision: https://reviews.llvm.org/D38607
llvm-svn: 315081
Unfortunately TableGen doesn't handle this yet:
Unable to deduce gMIR opcode to handle Src (which is a leaf).
Just add some temporary hand-written code to generate the proper MOVsr.
llvm-svn: 315071
The machine scheduler (before register allocation) is enabled by default for
SystemZ.
The SelectionDAG scheduling preference now becomes source order scheduling
(was regpressure).
Review: Ulrich Weigand
https://reviews.llvm.org/D37977
llvm-svn: 315063
Add extending loads and constant offset patterns
A bit more refactoring of the tablegen to make the patterns fairly nice and
uniform between the regular and atomic loads.
Differential Revision: https://reviews.llvm.org/D38523
llvm-svn: 315022
Summary: Make test robust enough to not fail due to CFG changes and re-enable for ARM/AArch64.
Reviewers: rovka, fhahn
Reviewed By: fhahn
Subscribers: fhahn, aemerson, rengolin, mcrosier, llvm-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D38590
llvm-svn: 315002
The inliner performs some kind of dead code elimination as it goes,
but there are cases that are not really caught by it. We might
at some point consider teaching the inliner about them, but it
is OK for now to run GlobalOpt + GlobalDCE in tandem as their
benefits generally outweight the cost, making the whole pipeline
faster.
This fixes PR34652.
Differential Revision: https://reviews.llvm.org/D38154
llvm-svn: 314997
The code which lowers BUILD_VECTOR of consecutive loads into a single vector
load doesn't update chains properly. As a result the vector load can be
reordered with the store to the same location.
The current code in EltsFromConsecutiveLoads only updates the chain following
the first load. The fix is to update the chains following all the loads
comprising the vector.
This is a fix for PR10114.
Reviewed By: niravd
Differential Revision: https://reviews.llvm.org/D38547
llvm-svn: 314988
It broke the Chromium / SQLite build; see PR34830.
> Summary:
> 1/ Operand folding during complex pattern matching for LEAs has been
> extended, such that it promotes Scale to accommodate similar operand
> appearing in the DAG.
> e.g.
> T1 = A + B
> T2 = T1 + 10
> T3 = T2 + A
> For above DAG rooted at T3, X86AddressMode will no look like
> Base = B , Index = A , Scale = 2 , Disp = 10
>
> 2/ During OptimizeLEAPass down the pipeline factorization is now performed over LEAs
> so that if there is an opportunity then complex LEAs (having 3 operands)
> could be factored out.
> e.g.
> leal 1(%rax,%rcx,1), %rdx
> leal 1(%rax,%rcx,2), %rcx
> will be factored as following
> leal 1(%rax,%rcx,1), %rdx
> leal (%rdx,%rcx) , %edx
>
> 3/ Aggressive operand folding for AM based selection for LEAs is sensitive to loops,
> thus avoiding creation of any complex LEAs within a loop.
>
> Reviewers: lsaba, RKSimon, craig.topper, qcolombet, jmolloy
>
> Reviewed By: lsaba
>
> Subscribers: jmolloy, spatel, igorb, llvm-commits
>
> Differential Revision: https://reviews.llvm.org/D35014
llvm-svn: 314919
This patch redefines the MOVSS/MOVSD instructions to take VR128 as its second input. This allows the MOVSS/SD->BLEND commute to work without requiring a COPY to be inserted.
This should fix PR33079
Overall this looks to be an improvement in the generated code. I haven't checked the EXPENSIVE_CHECKS build but I'll do that and update with results.
Differential Revision: https://reviews.llvm.org/D38449
llvm-svn: 314914
r314857 changed the CFG that resulted in the flaky test MachineBranchProb.ll to
fail the bots again. Marking it as unsupported for ARM/AArch64 again until we
find the cause.
llvm-svn: 314912
Previously, on long branches (relative jumps of >4 kB), an assertion
failure was hit, as AVRInstrInfo::insertIndirectBranch was not
implemented. Despite its name, it is called by the branch relaxator
for *all* unconditional jumps.
Patch by Thomas Backman.
llvm-svn: 314891
In some cases, the code generator attempts to generate instructions such as:
lddw r24, Y+63
which expands to:
ldd r24, Y+63
ldd r25, Y+64 # Oops! This is actually ld r25, Y in the binary
This commit limits the first offset to 62, and thus the second to 63.
It also updates some asserts in AVRExpandPseudoInsts.cpp, including for
INW and OUTW, which appear to be unused.
Patch by Thomas Backman.
llvm-svn: 314890
Summary:
1/ Operand folding during complex pattern matching for LEAs has been
extended, such that it promotes Scale to accommodate similar operand
appearing in the DAG.
e.g.
T1 = A + B
T2 = T1 + 10
T3 = T2 + A
For above DAG rooted at T3, X86AddressMode will no look like
Base = B , Index = A , Scale = 2 , Disp = 10
2/ During OptimizeLEAPass down the pipeline factorization is now performed over LEAs
so that if there is an opportunity then complex LEAs (having 3 operands)
could be factored out.
e.g.
leal 1(%rax,%rcx,1), %rdx
leal 1(%rax,%rcx,2), %rcx
will be factored as following
leal 1(%rax,%rcx,1), %rdx
leal (%rdx,%rcx) , %edx
3/ Aggressive operand folding for AM based selection for LEAs is sensitive to loops,
thus avoiding creation of any complex LEAs within a loop.
Reviewers: lsaba, RKSimon, craig.topper, qcolombet, jmolloy
Reviewed By: lsaba
Subscribers: jmolloy, spatel, igorb, llvm-commits
Differential Revision: https://reviews.llvm.org/D35014
llvm-svn: 314886
This time invoking llc with "-march=x86-64" in the testcase, so we don't assume
the default target is x86.
Summary:
If we have
%vreg0<def> = PHI %vreg2<undef>, <BB#0>, %vreg3, <BB#2>; GR32:%vreg0,%vreg2,%vreg3
%vreg3<def,tied1> = ADD32ri8 %vreg0<kill,tied0>, 1, %EFLAGS<imp-def>; GR32:%vreg3,%vreg0
then we can't just change %vreg0 into %vreg3, since %vreg2 is actually
undef. We would have to also copy the undef flag to be able to change the
register.
Instead we deal with this case like other cases where we can't just
replace the register: we insert a COPY. The code creating the COPY already
copied all flags from the PHI input, so the undef flag will be transferred
as it should.
Reviewers: kparzysz
Reviewed By: kparzysz
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38235
llvm-svn: 314882
Summary:
If we have
%vreg0<def> = PHI %vreg2<undef>, <BB#0>, %vreg3, <BB#2>; GR32:%vreg0,%vreg2,%vreg3
%vreg3<def,tied1> = ADD32ri8 %vreg0<kill,tied0>, 1, %EFLAGS<imp-def>; GR32:%vreg3,%vreg0
then we can't just change %vreg0 into %vreg3, since %vreg2 is actually
undef. We would have to also copy the undef flag to be able to change the
register.
Instead we deal with this case like other cases where we can't just
replace the register: we insert a COPY. The code creating the COPY already
copied all flags from the PHI input, so the undef flag will be transferred
as it should.
Reviewers: kparzysz
Reviewed By: kparzysz
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38235
llvm-svn: 314879
The previous version didn't work if the jump table base address didn't
fit in 32 bit, since it was encoded as an immediate offset. And in case
the jump table is encoded as 32 bit label differences, we need to
load and add them to the table base first.
This solves the first half of the issues mentioned in PR34720.
Also fix some of the errors pointed out by -verify-machineinstrs, by
using GR32_NOSPRegClass.
Differential Revision: https://reviews.llvm.org/D38333
llvm-svn: 314876
Summary:
After r308422 we defer optimizations that can destroy loop canonical forms to
LateSimplifyCFG. Running LateSimplifyCFG after expanding atomic operations
can exploit more control-flow opportunities.
Reviewers: mcrosier, t.p.northover, efriedma
Reviewed By: efriedma
Subscribers: aemerson, rengolin, javed.absar, llvm-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D38262
llvm-svn: 314857
This commit does two things. Firstly, it cleans up some of the benefit
calculation wrt outlined functions and candidates. Secondly, it fixes an
off-by-one bug in the cost model which was caused by the benefit value of
an OutlinedFunction and Candidate differing by 1. It updates the remarks test
to reflect this change.
llvm-svn: 314836
Summary:
For the amdpal OS type:
We write an AMDGPU_PAL_METADATA record in the .note section in the ELF
(or as an assembler directive). It contains key=value pairs of 32 bit
ints. It is a merge of metadata from codegen of the shaders, and
metadata provided by the frontend as _amdgpu_pal_metadata IR metadata.
Where both sources have a key=value with the same key, the two values
are ORed together.
This .note record is part of the amdpal ABI and will be documented in
docs/AMDGPUUsage.rst in a future commit.
Eventually the amdpal OS type will stop generating the .AMDGPU.config
section once the frontend has safely moved over to using the .note
records above instead of .AMDGPU.config.
Reviewers: arsenm, nhaehnle, dstuttard
Subscribers: kzhuravl, wdng, yaxunl, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D37753
llvm-svn: 314829
The dsp register class is an alias of the gpr register class, so
we have to define instructions for spilling and reloading.
Reviewers: atanasyan
Differential Revision: https://reviews.llvm.org/D38038
llvm-svn: 314798
If the upper bits of a truncation shuffle patterns have at least the minimum number of sign/zero bits on their inputs then we can safely use PACKSS/PACKUS as shuffles.
Partial fix for https://bugs.llvm.org/show_bug.cgi?id=34773
Differential Revision: https://reviews.llvm.org/D38472
llvm-svn: 314788
Summary:
Take the target's endianness into account when splitting the
debug information in DAGTypeLegalizer::SetExpandedInteger.
This patch fixes so that, for big-endian targets, the fragment
expression corresponding to the high part of a split integer
value is placed at offset 0, in order to correctly represent
the memory address order.
I have attached a PPC32 reproducer where the resulting DWARF
pieces for a 64-bit integer were incorrectly reversed.
Original patch was reverted due to using -stop-after=isel in
the test case (but that is only working when AMDGPU target
is included in the llc build). The test case has now been
updated to use -stop-before=expand-isel-pseudos instead.
Patch by: dstenb
Reviewers: JDevlieghere, aprantl, dblaikie
Reviewed By: JDevlieghere, aprantl, dblaikie
Subscribers: nemanjai
Differential Revision: https://reviews.llvm.org/D38172
llvm-svn: 314781
This makes sure the LSDA pointer isn't truncated to 32 bit.
Make LowerINTRINSIC_WO_CHAIN a member function instead of a static
function, so that it can use the getGlobalWrapperKind method.
This solves the second half of the issues mentioned in PR34720.
Differential Revision: https://reviews.llvm.org/D38343
llvm-svn: 314767
Issues addressed since original review:
- Avoid bug in regalloc greedy/machine verifier when forwarding to use
in an instruction that re-defines the same virtual register.
- Fixed bug when forwarding to use in EarlyClobber instruction slot.
- Fixed incorrect forwarding to register definitions that showed up in
explicit_uses() iterator (e.g. in INLINEASM).
- Moved removal of dead instructions found by
LiveIntervals::shrinkToUses() outside of loop iterating over
instructions to avoid instructions being deleted while pointed to by
iterator.
- Fixed ARMLoadStoreOptimizer bug exposed by this change in r311907.
- The pass no longer forwards COPYs to physical register uses, since
doing so can break code that implicitly relies on the physical
register number of the use.
- The pass no longer forwards COPYs to undef uses, since doing so
can break the machine verifier by creating LiveRanges that don't
end on a use (since the undef operand is not considered a use).
[MachineCopyPropagation] Extend pass to do COPY source forwarding
This change extends MachineCopyPropagation to do COPY source forwarding.
This change also extends the MachineCopyPropagation pass to be able to
be run during register allocation, after physical registers have been
assigned, but before the virtual registers have been re-written, which
allows it to remove virtual register COPY LiveIntervals that become dead
through the forwarding of all of their uses.
llvm-svn: 314729
These check lines are supposed to make sure the new d16
load instructions aren't used, but the expected instruction
name is a prefix of the incorrect instruction name.
llvm-svn: 314714
Summary: Also add support for some older Myriad CPUs that were missing.
Reviewers: jyknight
Subscribers: fedor.sergeev
Differential Revision: https://reviews.llvm.org/D37552
llvm-svn: 314705
Summary:
Take the target's endianness into account when splitting the
debug information in DAGTypeLegalizer::SetExpandedInteger.
This patch fixes so that, for big-endian targets, the fragment
expression corresponding to the high part of a split integer
value is placed at offset 0, in order to correctly represent
the memory address order.
I have attached a PPC32 reproducer where the resulting DWARF
pieces for a 64-bit integer were incorrectly reversed.
Patch by: dstenb
Reviewers: JDevlieghere, aprantl, dblaikie
Reviewed By: JDevlieghere, aprantl, dblaikie
Subscribers: nemanjai
Differential Revision: https://reviews.llvm.org/D38172
llvm-svn: 314666
This patch add a support of ISD::ZERO_EXTEND in PPCDAGToDAGISel::tryBitPermutation to increase the opportunity to use rotate-and-mask by reordering ZEXT and ANDI.
Since tryBitPermutation stops analyzing nodes if it hits a ZEXT node while traversing SDNodes, we want to avoid ZEXT between two nodes that can be folded into a rotate-and-mask instruction.
For example, we allow these nodes
t9: i32 = add t7, Constant:i32<1>
t11: i32 = and t9, Constant:i32<255>
t12: i64 = zero_extend t11
t14: i64 = shl t12, Constant:i64<2>
to be folded into a rotate-and-mask instruction.
Such case often happens in array accesses with logical AND operation in the index, e.g. array[i & 0xFF];
Differential Revision: https://reviews.llvm.org/D37514
llvm-svn: 314655
I continue to support different VF interleaved and in this pass for this patch,
I added the vf64 stride3 support for both load and store.
I also added support fot the stride4 store.
Reviewers:
1. zvi
2. dorit
3. igorb
4. guyblank
Differential Revision: https://reviews.llvm.org/D37687
Change-Id: I3d238efedf217d1768b348d710de1efa2f19d27b
llvm-svn: 314651
If the two instructions being compared for equivalence have corresponding operands
that are integer constants, then check their values to determine equivalence.
Patch by Suyog Sarda!
llvm-svn: 314642
Summary:
Intel documentation shows the memory operand as the first operand. But we currently treat it as the second operand. Conceptually the order doesn't matter since it doesn't write memory. We have aliases to parse with the operands in either order and the isel matching is commutable.
For the register®ister form order does matter for the assembly parser. PR22995 was previously filed and fixed by changing the register®ister form from MRMSrcReg to MRMDestReg to match gas. Ideally the memory form should match by using MRMDestMem.
I believe this supercedes D38025 which was trying to switch the register®ister form back to pre-PR22995.
Reviewers: aymanmus, RKSimon, zvi
Reviewed By: aymanmus
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38120
llvm-svn: 314639
NFC.
Added code gen regression tests for avx512 instructions scheduling called avx512-schedule.ll and
avx512-shuffle-schedule.ll.
This patch is in preparation of a larger patch of adding all SKX instruction scheduling and therefore
the scheduling for the avx512 instructions are still missing.
Reviewers: zvi, delena, RKSimon, igorb
Differential Revision: https://reviews.llvm.org/D38035
Change-Id: I792762763127a921b9e13684b58af03646536533
llvm-svn: 314594
Implemented by splitting into two v32i8 mulhu/mulhs and concatenating the results.
Differential Revision: https://reviews.llvm.org/D38307
llvm-svn: 314584
We have a single library build without relaxation options.
When inlined library functions remove fast math attributes
from the functions they are integrated into.
This patch sets relaxation attributes on the functions after
linking provided corresponding relaxation options are given.
Math instructions inside the inlined functions remain to have
no fast flags, but inlining does not prevent fast math
transformations of a surrounding caller code anymore.
Differential Revision: https://reviews.llvm.org/D38325
llvm-svn: 314568
Currently expandUnalignedLoad/Store uses place holder pointer info for temporary memory operand
in stack, which does not have correct address space. This causes unaligned private double16 load/store to be
lowered to flat_load instead of buffer_load for amdgcn target.
This fixes failures of OpenCL conformance test basic/vload_private/vstore_private on target amdgcn---amdgizcl.
Differential Revision: https://reviews.llvm.org/D35361
llvm-svn: 314566
The hardware will only forward EXEC_LO; the high 32 bits will be zero.
Additionally, inline constants do not work. At least,
v_addc_u32_e64 v0, vcc, v0, v1, -1
which could conceivably be used to combine (v0 + v1 + 1) into a single
instruction, acts as if all carry-in bits are zero.
The llvm.amdgcn.ps.live test is adjusted; it would be nice to combine
s_mov_b64 s[0:1], exec
v_cndmask_b32_e64 v0, v1, v2, s[0:1]
into
v_mov_b32 v0, v3
but it's not particularly high priority.
Fixes dEQP-GLES31.functional.shaders.helper_invocation.value.*
llvm-svn: 314522
Implement shouldCoalesce() to help regalloc avoid running out of GR128
registers.
If a COPY involving a subreg of a GR128 is coalesced, the live range of the
GR128 virtual register will be extended. If this happens where there are
enough phys-reg clobbers present, regalloc will run out of registers (if
there is not a single GR128 allocatable register available).
This patch tries to allow coalescing only when it can prove that this will be
safe by checking the (local) interval in question.
Review: Ulrich Weigand, Quentin Colombet
https://reviews.llvm.org/D37899https://bugs.llvm.org/show_bug.cgi?id=34610
llvm-svn: 314516
Fix nested callseq* nodes by moving callseq_start after the
arguments calculation to temporary registers, so that callseq* nodes
in resulting DAG are linear.
Recommitting r314497. This version does not contain test which fails
when compiler is not build in debug mode.
Differential Revision: https://reviews.llvm.org/D37328
llvm-svn: 314507
Added additional tests for vector multiplications with multipliers that are:
* powers of 2 displaced by 1,
* product of a power of 2 displaced by one with another power of 2.
Patch by @pacxx (Michael Haidl)
Differential Revision: https://reviews.llvm.org/D38350
llvm-svn: 314504
Summary:
This commit adds comments on how the AMDPAL OS type overloads the
existing AMDGPU_ calling conventions used by Mesa, and adds a couple of
new ones.
Reviewers: arsenm, nhaehnle, dstuttard
Subscribers: mehdi_amini, kzhuravl, wdng, yaxunl, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D37752
llvm-svn: 314502
Summary:
Added support for scratch (including spilling) for OS type amdpal:
generates code to set up the scratch descriptor if it is needed.
With amdpal, the scratch resource descriptor is loaded from offset 0 of
the global information table. The low 32 bits of the address of the
global information table is passed in s0.
Added amdgpu-git-ptr-high function attribute to hard-wire the high 32
bits of the address of the global information table. If the function
attribute is not specified, or is 0xffffffff, then the backend generates
code to use the high 32 bits of pc.
The documentation for the AMDPAL ABI will be added in a later commit.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye
Differential Revision: https://reviews.llvm.org/D37483
llvm-svn: 314501
Summary:
This operating system type represents the AMDGPU PAL runtime, and will
be required by the AMDGPU backend in order to generate correct code for
this runtime.
Currently it generates the same code as not specifying an OS at all.
That will change in future commits.
Patch from Tim Corringham.
Subscribers: arsenm, nhaehnle
Differential Revision: https://reviews.llvm.org/D37380
llvm-svn: 314500
Fix nested callseq* nodes by moving callseq_start after the
arguments calculation to temporary registers, so that callseq* nodes
in resulting DAG are linear.
Differential Revision: https://reviews.llvm.org/D37328
llvm-svn: 314497
Summary:
X86ISelDAGToDAG tries to analyze ANDs compared with 0 to optimize to narrower immediates using subregisters.
I don't think we should be optimizing to 16-bit test instructions. It goes against our normal behavior of promoting i16 operations to i32. It only saves one byte due to the need to add a 0x66 prefix. I think it would also be subject to a length changing prefix penalty in the decoders on Intel CPUs.
Reviewers: RKSimon, zvi, spatel
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38273
llvm-svn: 314474
LR is an untypical callee saved register in that it is restored into a
different register (PC) and thus does not live-out of the return block.
This case requires the `Restored` flag in CalleeSavedInfo to be cleared.
This fixes a number of cases where this wasn't handled correctly yet.
llvm-svn: 314471
The expensive-checks build bot found a problem with the r314428 commit:
if CC is live after a ATOMIC_CMP_SWAPW instruction, it needs to be
marked as live-in to the block after the loop the pseudo gets expanded
to. This actually fixes a code-gen bug as well, since if the CC isn't
live, the CR and JLH are merged to a CRJLH which doesn't actually set
the condition code any more.
llvm-svn: 314465
If we have BWI, we can truncate in a much simpler way by using vpmovwb. This even works without VLX by using the wider zmm->ymm truncate with a subvector extract.
Differential Revision: https://reviews.llvm.org/D38375
llvm-svn: 314457
In setupEntryBlockAndCallSites in CodeGen/SjLjEHPrepare.cpp,
we fetch and store the actual frame pointer, but on return via
the longjmp intrinsic, it always was restored into the r7 variable.
On windows, the frame pointer should be restored into r11 instead of r7.
On Darwin (where sjlj exception handling is used by default), the frame
pointer is always r7, both in arm and thumb mode, and likewise, on
windows, the frame pointer always is r11.
On linux however, if sjlj exception handling is enabled (which it isn't
by default), libcxxabi and the user code can be built in differing modes
using different registers as frame pointer. Therefore, when restoring
registers on a platform where we don't always use the same register
depending on code mode, restore both r7 and r11.
Differential Revision: https://reviews.llvm.org/D38253
llvm-svn: 314451
Yonghong Song