This reverts most of r274613 (AKA r274626) and its follow-ups (r276347, r277289),
due to miscompiles in the test suite. The FastISel change was left in, because
it apparently fixes an unrelated issue.
(Recommit of r279782 which was broken due to a bad merge.)
This fixes 4 out of the 5 test failures in PR29112.
llvm-svn: 279788
This reverts most of r274613 and its follow-ups (r276347, r277289), due to
miscompiles in the test suite. The FastISel change was left in, because it
apparently fixes an unrelated issue.
This fixes 4 out of the 5 test failures in PR29112.
llvm-svn: 279782
Fix VPAVG detection to require AVX512BW, not AVX512F for 512-bit widths,
and change associated asserts to assert in the right direction...
This fixes PR29111.
llvm-svn: 279755
Rename AllVRegsAllocated to NoVRegs. This avoids the connotation of
running after register and simply describes that no vregs are used in
a machine function. With that we can simply compute the property and do
not need to dump/parse it in .mir files.
Differential Revision: http://reviews.llvm.org/D23850
llvm-svn: 279698
tracksSubRegLiveness only depends on the Subtarget and a cl::opt, there
is not need to change it or save/parse it in a .mir file.
Make the field const and move the initialization LiveIntervalAnalysis to the
MachineRegisterInfo constructor. Also cleanup some code and fix some
instances which better use MachineRegisterInfo::subRegLivenessEnabled() instead
of TargetSubtargetInfo::enableSubRegLiveness().
llvm-svn: 279676
These are no different in load behaviour to the existing ADD/SUB/MUL/DIV scalar ops but were missing from isNonFoldablePartialRegisterLoad
llvm-svn: 279652
Includes adding more general support for the pattern: VZEXT_MOVL(VZEXT_LOAD(ptr)) -> VZEXT_LOAD(ptr)
This has unearthed a couple of latent poor codegen issues (MINSS/MAXSS scalar load folding and MOVDDUP/BROADCAST load folding patterns), which will be fixed shortly.
Its also reduced a couple of tests so that they no longer reach the instruction threshold necessary to be combined to PSHUFB (see PR26183).
llvm-svn: 279646
Re-apply this patch, hopefully I will get away without any warnings
in the constructor now.
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279602
Specifying isSSA is an extra line at best and results in invalid MI at
worst. Compute the value instead.
Differential Revision: http://reviews.llvm.org/D22722
llvm-svn: 279600
Re-apply this commit with the deletion of a MachineFunction delegated to
a separate pass to avoid use after free when doing this directly in
AsmPrinter.
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279564
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279502
As discussed on PR26491, we are missing the opportunity to make use of the smaller MOVHLPS instruction because we set both arguments of a SHUFPD when using it to lower a single input shuffle.
This patch sets the lowered argument to UNDEF if that shuffle element is undefined. This in turn makes it easier for target shuffle combining to decode UNDEF shuffle elements, allowing combines to MOVHLPS to occur.
A fix to match against MOVHPD stores was necessary as well.
This builds on the improved MOVLHPS/MOVHLPS lowering and memory folding support added in D16956
Adding similar support for SHUFPS will have to wait until have better support for target combining of binary shuffles.
Differential Revision: https://reviews.llvm.org/D23027
llvm-svn: 279430
In some cases, FastIsel was emitting TEST instruction with K reg input, which is illegal.
Changed to using KORTEST when dealing with K regs.
Differential Revision: https://reviews.llvm.org/D23163
llvm-svn: 279393
This doesn't change tests codegen as we already combined to blend+zero which is what we lower VZEXT_MOVL to on SSE41+ targets, but it does put us in a better position when we improve shuffling for optsize.
llvm-svn: 279273
INSERTPS doesn't fit well with our shuffle mask canonicalization, so we need to attempt both the original mask and the commuted mask to more likely get a match
llvm-svn: 279230
Without the synthesized reference to a symbol in the xray_instr_map,
linker section garbage collection will helpfully remove the whole
xray_instr_map section from the final executable (or archive). This will
cause the runtime to not be able to identify the sleds and hot-patch the
calls/jumps into the runtime trampolines.
This change adds a reference from the text section at the end of the
function to keep around the associated xray_instr_map section as well.
We also make sure that we catch this reference in the test.
Reviewers: chandlerc, echristo, majnemer, mehdi_amini
Subscribers: mehdi_amini, llvm-commits, dberris
Differential Revision: https://reviews.llvm.org/D23398
llvm-svn: 279204
This is a fix for https://llvm.org/bugs/show_bug.cgi?id=29010
Root cause of the bug is that the register class of the machine instruction operand does not fully reflect if this registers that can be allocated.
Both for i386 and x86_64 the operand's register class is VR128RegClass and thus contains xmm0-xmm15, though in i386 we can only use xmm0-xmm8.
In order to get the actual allocable registers of the class we need to use RegisterClassInfo.
Differential Revision: https://reviews.llvm.org/D23613
llvm-svn: 278954
This patch handles 64-bit constants which can be encoded as 32-bit immediates.
It extends the functionality added by https://reviews.llvm.org/D11363 for 32-bit constants to 64-bit constants.
Patch by Sunita Marathe!
Differential Revision: https://reviews.llvm.org/D23391
llvm-svn: 278857
Do not reorder and move up a loop latch block before a loop header
when optimising for size because this will generate an extra
unconditional branch.
Differential Revision: https://reviews.llvm.org/D22521
llvm-svn: 278840
If a loop is not rotated (for example when optimizing for size), the latch is not the backedge. If we promote an expression to post-inc form, we not only increase register pressure and add a COPY for that IV expression but for all IVs!
Motivating testcase:
void f(float *a, float *b, float *c, int n) {
while (n-- > 0)
*c++ = *a++ + *b++;
}
It's imperative that the pointer increments be located in the latch block and not the header block; if not, we cannot use post-increment loads and stores and we have to keep both the post-inc and pre-inc values around until the end of the latch which bloats register usage.
llvm-svn: 278658
1. Use shuffle to insert element i1 into vector. The previous implementation was incorrect ( dest_bit OR src_bit , it doesn't clear the bit if src_bit=0 )
2. Improve shuffle i1 vector, use CVT2MASK if supported instead TRUNCATE.
Differential Revision: http://reviews.llvm.org/D23347
llvm-svn: 278623
Currently X86ISelLowering has a similar transformation for sexts:
sext(add_nsw(x, C)) --> add(sext(x), C_sext)
In this change I extend this code to handle zexts as well.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D23359
llvm-svn: 278520
The PALIGNR target shuffle decode was not taking into account that DecodePALIGNRMask (rather oddly) expects the operands to be in reverse order, nor was it detecting unary patterns, causing combines to combine with the incorrect input.
The cgbuiltin, auto upgrade and instruction comments code correctly swap the operands so are not affected.
llvm-svn: 278494
"insert_subreg, subreg_to_reg, and reg_sequence" instructions' after
adjusting some unittest checks.
This is to solve PR28852. The restriction was added at 2010 to make better register
coalescing. We assumed that it was not necessary any more. Testing results on x86
supported the assumption.
We will look closely to any performance impact it will bring and will be prepared
to help analyzing performance problem found on other architectures.
Differential Revision: https://reviews.llvm.org/D23210
llvm-svn: 278466
To fix PR28014, this patch restricts tail merging to blocks that belong to the
same loop after MBP.
Differential Revision: https://reviews.llvm.org/D23191
llvm-svn: 278463
This helped to improved memory-folding and register coalescing optimizations.
Also, this patch fixed the tracker #17229.
Reviewer: Craig Topper.
Differential Revision: https://reviews.llvm.org/D23108
llvm-svn: 278431
subreg_to_reg, and reg_sequence" instructions.
This is to solve PR28852. The restriction was added at 2010 to make better register
coalescing. We assumed that it was not necessary any more. Testing results on x86
supported the assumption.
We will look closely to any performance impact it will bring and will be prepared
to help analyzing performance problem found on other architectures.
Differential Revision: https://reviews.llvm.org/D23210
llvm-svn: 278384
The previous implementation (not custom) doesn't enforce zeroing off upper bits. The assumption is that i1 PRODUCER (truncate and extractelement) must zero all upper bits, so i1 CONSUMER instructions ( test, zext, save, etc) can be done without additional zeroing.
Make extractelement i1 lowering custom for all vector i1.
Differential Revision: http://reviews.llvm.org/D23246
llvm-svn: 278328
This patch helps avoid false dependencies on undef registers by updating the machine instructions' undef operand to use a register that the instruction is truly dependent on, or use a register with clearance higher than Pref.
Pseudo example:
loop:
xmm0 = ...
xmm1 = vcvtsi2sdl eax, xmm0<undef>
... = inst xmm0
jmp loop
In this example, selecting xmm0 as the undef register creates false dependency between loop iterations.
This false dependency cannot be solved by inserting an xor before vcvtsi2sdl because xmm0 is alive at the point of the vcvtsi2sdl instruction.
Selecting a different register instead of xmm0, especially a register that is not used in the loop, will eliminate this problem.
Differential Revision: https://reviews.llvm.org/D22466
llvm-svn: 278321
If the input vector to INSERT_SUBVECTOR is another INSERT_SUBVECTOR, and this inserted subvector replaces the last insertion, then insert into the common source vector.
i.e.
INSERT_SUBVECTOR( INSERT_SUBVECTOR( Vec, SubOld, Idx ), SubNew, Idx ) --> INSERT_SUBVECTOR( Vec, SubNew, Idx )
Differential Revision: https://reviews.llvm.org/D23330
llvm-svn: 278211
A UD2 might make its way into the program via a call to @llvm.trap.
Obviously, calls are not terminators. However, we modeled the X86
instruction, UD2, as a terminator. Later on, this confuses the epilogue
insertion machinery which results in the epilogue getting inserted
before the UD2. For some platforms, like x64, the result is a
violation of the ABI.
Instead, model UD2/UD2B as a side effecting instruction which may
observe memory.
llvm-svn: 278144
As detailed on D22726, much of the shift combining code assume constant values will fit into a uint64_t value and calls ConstantSDNode::getZExtValue where it probably shouldn't (leading to asserts). Using APInt directly avoids this problem but we encounter other assertions if we attempt to compare/operate on 2 APInt of different bitwidths.
This patch adds a helper function to ensure that 2 APInt values are zero extended as required so that they can be safely used together. I've only added an initial example use for this to the '(SHIFT (SHIFT x, c1), c2) --> (SHIFT x, (ADD c1, c2))' combines. Further cases can easily be added as required.
Differential Revision: https://reviews.llvm.org/D23007
llvm-svn: 278141
This makes a trivial change in the emission of the per-function XRay
tables, and makes sure that the xray_instr_map section does show up in
the object file.
llvm-svn: 278113
We only had partial memory folding support for the intrinsic definitions, and (as noted on PR27481) was causing FR32/FR64/VR128 mismatch errors with the machine verifier.
This patch adds missing memory folding support for both intrinsics and the ffloor/fnearbyint/fceil/frint/ftrunc patterns and in doing so fixes the failing machine verifier stack folding tests from PR27481.
Differential Revision: https://reviews.llvm.org/D23276
llvm-svn: 278106
This reverts commit r278048. Something changed between the last time I
built this--it takes awhile on my ridiculously slow and ancient
computer--and now that broke this.
llvm-svn: 278053
Summary:
Based on two patches by Michael Mueller.
This is a target attribute that causes a function marked with it to be
emitted as "hotpatchable". This particular mechanism was originally
devised by Microsoft for patching their binaries (which they are
constantly updating to stay ahead of crackers, script kiddies, and other
ne'er-do-wells on the Internet), but is now commonly abused by Windows
programs to hook API functions.
This mechanism is target-specific. For x86, a two-byte no-op instruction
is emitted at the function's entry point; the entry point must be
immediately preceded by 64 (32-bit) or 128 (64-bit) bytes of padding.
This padding is where the patch code is written. The two byte no-op is
then overwritten with a short jump into this code. The no-op is usually
a `movl %edi, %edi` instruction; this is used as a magic value
indicating that this is a hotpatchable function.
Reviewers: majnemer, sanjoy, rnk
Subscribers: dberris, llvm-commits
Differential Revision: https://reviews.llvm.org/D19908
llvm-svn: 278048
Optimized lowering of BITCAST node. The BITCAST node can be replaced with COPY_TO_REG instead of KMOV.
It allows to suppress two opposite BITCAST operations and avoid redundant "movs".
Differential Revision: https://reviews.llvm.org/D23247
llvm-svn: 277958
This is a new test that should explore a current suboptimal sequence in passing values between cmp and kor intrinsics.
The code will be optimized in an upcoming patch.
Submitted bug here:
https://llvm.org/bugs/show_bug.cgi?id=28839
llvm-svn: 277954
Assuming SSE2 is available then we can safely commute between these, removing some unnecessary register moves and improving memory folding opportunities.
VEX encoded versions don't benefit so I haven't added support to them.
llvm-svn: 277930
On modern Intel processors hardware SQRT in many cases is faster than RSQRT
followed by Newton-Raphson refinement. The patch introduces a simple heuristic
to choose between hardware SQRT instruction and Newton-Raphson software
estimation.
The patch treats scalars and vectors differently. The heuristic is that for
scalars the compiler should optimize for latency while for vectors it should
optimize for throughput. It is based on the assumption that throughput bound
code is likely to be vectorized.
Basically, the patch disables scalar NR for big cores and disables NR completely
for Skylake. Firstly, scalar SQRT has shorter latency than NR code in big cores.
Secondly, vector SQRT has been greatly improved in Skylake and has better
throughput compared to NR.
Differential Revision: https://reviews.llvm.org/D21379
llvm-svn: 277725
This should ensure that we can atomically write two bytes (on top of the
retq and the one past it) and have those two bytes not straddle cache
lines.
We also move the label past the alignment instruction so that we can refer
to the actual first instruction, as opposed to potential padding before the
aligned instruction.
Update the tests to allow us to reflect the new order of assembly.
Reviewers: rSerge, echristo, majnemer
Subscribers: llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D23101
llvm-svn: 277701
We currently only support combining target shuffles that consist of a single source input (plus elements known to be undef/zero).
This patch generalizes the recursive combining of the target shuffle to collect all the inputs, merging any duplicates along the way, into a full set of src ops and its shuffle mask.
We uncover a number of cases where we have failed to combine a unary shuffle because the input has been duplicated and separated during lowering.
This will allow us to combine to 2-input shuffles in a future patch.
Differential Revision: https://reviews.llvm.org/D22859
llvm-svn: 277631
Summary:
We also add a test to show what currently happens when we create a
section per function and emit an xray_instr_map. This illustrates the
relationship (or lack thereof) between the per-function section and the
xray_instr_map section.
We also change the code generation slightly so that we don't always
create group sections, but rather only do so if a function where the
table is associated with is in a group.
Also in this change:
- Remove the "merge" flag on the xray_instr_map section.
- Test that we're generating the right table for comdat and non-comdat functions.
Reviewers: echristo, majnemer
Subscribers: llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D23104
llvm-svn: 277580
We used to combine "sext(setcc x, y, cc) -> (select (setcc x, y, cc), -1, 0)"
Instead, we should combine to (select (setcc x, y, cc), T, 0) where the value
of T is 1 or -1, depending on the type of the setcc, and getBooleanContents()
for the type if it is not i1.
This fixes PR28504.
llvm-svn: 277371
As discussed on PR14593, this patch adds support for lowering to SHLD/SHRD from the patterns generated by DAGTypeLegalizer::ExpandShiftWithKnownAmountBit.
Differential Revision: https://reviews.llvm.org/D23000
llvm-svn: 277299
Up until now, we only had code to match PSADBW patterns that look like what
comes out of the loop vectorizer - a partial reduction inside the loop body
that gets fed into a horizontal operation in a different basic block.
This adds support for straight-line patterns, like those generated by the
SLP vectorizer.
Differential Revision: https://reviews.llvm.org/D22889
llvm-svn: 277219
Support for lowering to VBROADCASTF128 etc. in D22460 was not correctly ensuring that the only users of the 128-bit vector load were the insertions of the vector into the lower/upper subvectors.
llvm-svn: 277214
Patch by Sunita Marathe
Third try, now following fixes to MSan to handle mempcy in such a way that this commit won't break the MSan buildbots. (Thanks, Evegenii!)
llvm-svn: 277189
The following pattern was being layed out poorly:
A
/ \
B C
/ \ / \
D E ? (Doesn't matter)
Where A->B is far more likely than A->C, and prob(B->D) = prob(B->E)
The current algorithm gives:
A,B,C,E (D goes on worklist)
It does this even if C has a frequency count of 0. This patch
adjusts the layout calculation so that if freq(B->E) >> freq(C->E)
then we go ahead and layout E rather than C. Fallthrough half the time
is better than fallthrough never, or fallthrough very rarely. The
resulting layout is:
A,B,E, (C and D are in a worklist)
llvm-svn: 277187
We currently default to using either generic shuffles or MASK+PACKUS/PACKSS to truncate all integer vectors. For vector comparisons, we know that the result will be either all or zero bits in every element, which can be efficiently truncated by directly using PACKSS to repeatedly halve the size of each element.
Due to the limited input values (-1 or 0) we don't need to account for vector element size, so for simplicity we just use the PACKSS(vXi16,vXi16) implementation in all cases. Additionally for AVX2 PACKSS of 256bit data we must perform a PERMQ shuffle to reorder the data into the correct order. I did investigate performing a single shuffle after all the PACKSS calls but the need to cross 128bit lanes makes this difficult to achieve efficiently.
We avoid performing this on AVX512 as it should have better alternative truncation instructions.
Differential Revision: https://reviews.llvm.org/D22814
llvm-svn: 277132
I'm not convinced the patterns for the rm_Int was correct anyway. It had a tied source that should't exist for the unmasked version. The load form of MOVSS always zeros the most significant bits. I've left the patterns off the masked load instructions as I'm not sure what the correct pattern should be and we don't have any tests currently. Nor do we implement masked scalar load intrinsics in clang currently.
llvm-svn: 277098
Michael Kuperstein