Add codegen for VSX word extract conversion from signed/unsigned to single/double
precision.
For UINT_TO_FP:
Extract word unsigned and convert to float was implemented in https://reviews.llvm.org/D20239.
Here we will add the missing extract integer and conversion to double. This
utilizes the new P9 instruction xxextractuw to extracting an integer element
when the result will be converted to double thereby saving 2 direct moves
(VSR <-> GPR).
For SINT_TO_FP:
We will implement the following sequence which will also reduce the number of
instructions by saving 2 direct moves.
v4i32->f32:
xxspltw
xvcvsxwsp
xscvspdpn
v4i32->f64:
xxspltw
xvcvsxwdp
Differential Revision: https://reviews.llvm.org/D35859
llvm-svn: 310866
introduce a miscompile bug.
There appears to be a bug where the generated code to extract the sign
bit doesn't work correctly for 32-bit inputs. I've replied to the
original commit pointing out the problem. I think I see by inspection
(and reading the manual for PPC) how to fix this, but I can't be 100%
confident and I also don't know what the best way to test this is.
Currently it seems nearly impossible to get the backend to hit this code
path, but the patch autohr is likely in a better position to craft such
test cases than I am, and based on where the bug is it should be easily
done.
Original commit message for r310346:
"""
[PowerPC] Eliminate compares - add i32 sext/zext handling for SETLE/SETGE
Adds handling for SETLE/SETGE comparisons on i32 values. Furthermore, it
adds the handling for the special case where RHS == 0.
Differential Revision: https://reviews.llvm.org/D34048
"""
llvm-svn: 310809
Summary:
This fixes PR32721 in IfConvertTriangle and possible similar problems in
IfConvertSimple, IfConvertDiamond and IfConvertForkedDiamond.
In PR32721 we had a triangle
EBB
| \
| |
| TBB
| /
FBB
where FBB didn't have any successors at all since it ended with an
unconditional return. Then TBB and FBB were be merged into EBB, but EBB
would still keep its successors, and the use of analyzeBranch and
CorrectExtraCFGEdges wouldn't help to remove them since the return
instruction is not analyzable (at least not on ARM).
The edge updating code and branch probability updating code is now pushed
into MergeBlocks() which allows us to share the same update logic between
more callsites. This lets us remove several dependencies on analyzeBranch
and completely eliminate RemoveExtraEdges.
One thing that showed up with this patch was that IfConversion sometimes
left a successor with 0% probability even if there was no branch or
fallthrough to the successor.
One such example from the test case ifcvt_bad_zero_prob_succ.mir. The
indirect branch tBRIND can only jump to bb.1, but without the patch we
got:
bb.0:
successors: %bb.1(0x80000000)
bb.1:
successors: %bb.1(0x80000000), %bb.2(0x00000000)
tBRIND %r1, 1, %cpsr
B %bb.1
bb.2:
There is no way to jump from bb.1 to bb2, but still there is a 0% edge
from bb.1 to bb.2.
With the patch applied we instead get the expected:
bb.0:
successors: %bb.1(0x80000000)
bb.1:
successors: %bb.1(0x80000000)
tBRIND %r1, 1, %cpsr
B %bb.1
Since bb.2 had no predecessor at all, it was removed.
Several testcases had to be updated due to this since the removed
successor made the "Branch Probability Basic Block Placement" pass
sometimes place blocks in a different order.
Finally added a couple of new test cases:
* PR32721_ifcvt_triangle_unanalyzable.mir:
Regression test for the original problem dexcribed in PR 32721.
* ifcvt_triangleWoCvtToNextEdge.mir:
Regression test for problem that caused a revert of my first attempt
to solve PR 32721.
* ifcvt_simple_bad_zero_prob_succ.mir:
Test case showing the problem where a wrong successor with 0% probability
was previously left.
* ifcvt_[diamond|forked_diamond|simple]_unanalyzable.mir
Very simple test cases for the simple and (forked) diamond cases
involving unanalyzable branches that can be nice to have as a base if
wanting to write more complicated tests.
Reviewers: iteratee, MatzeB, grosser, kparzysz
Reviewed By: kparzysz
Subscribers: kbarton, davide, aemerson, nemanjai, javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D34099
llvm-svn: 310697
We've implemented a 1-byte splat using XXSPLTISB on P9. However, LLVM will
produce a 1-byte splat even for wider element BUILD_VECTOR nodes. This patch
prevents crashing in that situation.
Differential Revision: https://reviews.llvm.org/D35650
llvm-svn: 310358
Adds handling for SETLE/SETGE comparisons on i32 values. Furthermore, it adds
the handling for the special case where RHS == 0.
Differential Revision: https://reviews.llvm.org/D34048
llvm-svn: 310346
Power 9 has instructions to do absolute difference (VABSDUB, VABSDUH, VABSDUW)
for byte, halfword and word. We should take advantage of these.
Differential Revision: https://reviews.llvm.org/D34684
llvm-svn: 309876
`llc -march` is problematic because it only switches the target
architecture, but leaves the operating system unchanged. This
occasionally leads to indeterministic tests because the OS from
LLVM_DEFAULT_TARGET_TRIPLE is used.
However we can simply always use `llc -mtriple` instead. This changes
all the tests to do this to avoid people using -march when they copy and
paste parts of tests.
This patch:
- Removes -march if the .ll file already has a matching `target triple`
directive or -mtriple argument.
- In all other cases changes -march=ppc32/-march=ppc64 to
-mtriple=ppc32--/-mtriple=ppc64--
See also the discussion in https://reviews.llvm.org/D35287
llvm-svn: 309754
As noted in the code comment, transforming this in the other direction might require
a separate transform here in CGP given the block-at-a-time DAG constraint.
Besides that theoretical motivation, there are 2 practical motivations for the
subtract-of-cmps form:
1. The codegen for both x86 and PPC is better for this IR (though PPC could be better still).
There is discussion about canonicalizing IR to the select form
( http://lists.llvm.org/pipermail/llvm-dev/2017-July/114885.html ),
so we probably need to add DAG transforms for those patterns anyway, but this improves the
memcmp output without waiting for that step.
2. If we allow vector-sized chunks for the load and compare, x86 is better prepared to convert
that to optimal code when using subtract-of-cmps, so another prerequisite patch is avoided
if we choose to enable that.
Differential Revision: https://reviews.llvm.org/D34904
llvm-svn: 309597
In optimizeCompareInstr, a compare instruction is eliminated by using a record form instruction if possible.
If the branch instruction that uses the result of the compare has a static branch hint, the optimization does not happen.
This patch makes this optimization happen regardless of the branch hint by splitting branch hint and branch condition before checking the predicate to identify the possible optimizations.
Differential Revision: https://reviews.llvm.org/D35801
llvm-svn: 309255
This patch just adds printing of CR bit registers in a more human-readable
form akin to that used by the GNU binutils.
Differential Revision: https://reviews.llvm.org/D31494
llvm-svn: 309001
As outlined in the PR, we didn't ensure that displacements for DQ-Form
instructions are multiples of 16. Since the instruction encoding encodes
a quad-word displacement, a sub-16 byte displacement is meaningless and
ends up being encoded incorrectly.
Fixes https://bugs.llvm.org/show_bug.cgi?id=33671.
Differential Revision: https://reviews.llvm.org/D35007
llvm-svn: 307934
OpenCL 2.0 introduces the notion of memory scopes in atomic operations to
global and local memory. These scopes restrict how synchronization is
achieved, which can result in improved performance.
This change extends existing notion of synchronization scopes in LLVM to
support arbitrary scopes expressed as target-specific strings, in addition to
the already defined scopes (single thread, system).
The LLVM IR and MIR syntax for expressing synchronization scopes has changed
to use *syncscope("<scope>")*, where <scope> can be "singlethread" (this
replaces *singlethread* keyword), or a target-specific name. As before, if
the scope is not specified, it defaults to CrossThread/System scope.
Implementation details:
- Mapping from synchronization scope name/string to synchronization scope id
is stored in LLVM context;
- CrossThread/System and SingleThread scopes are pre-defined to efficiently
check for known scopes without comparing strings;
- Synchronization scope names are stored in SYNC_SCOPE_NAMES_BLOCK in
the bitcode.
Differential Revision: https://reviews.llvm.org/D21723
llvm-svn: 307722
1. The available program storage region of the red zone to compilers is 288
bytes rather than 244 bytes.
2. The formula for negative number alignment calculation should be
y = x & ~(n-1) rather than y = (x + (n-1)) & ~(n-1).
Differential Revision: https://reviews.llvm.org/D34337
llvm-svn: 307672
For this example:
float test (int *arr) {
return arr[2];
}
We currently generate the following code:
li r4, 8
lxsiwax f0, r3, r4
xscvsxdsp f1, f0
With this patch, we will now generate:
addi r3, r3, 8
lxsiwax f0, 0, r3
xscvsxdsp f1, f0
Originally reported in: https://bugs.llvm.org/show_bug.cgi?id=27204
Differential Revision: https://reviews.llvm.org/D35027
llvm-svn: 307553
This patch adds the exploitation for new power 9 instructions which extract
variable elements from vectors:
VEXTUBLX
VEXTUBRX
VEXTUHLX
VEXTUHRX
VEXTUWLX
VEXTUWRX
Differential Revision: https://reviews.llvm.org/D34032
Commit on behalf of Zaara Syeda (syzaara@ca.ibm.com)
llvm-svn: 307174
This patch adds on to the exploitation added by https://reviews.llvm.org/D33510.
This now catches build vector nodes where the inputs are coming from sign
extended vector extract elements where the indices used by the vector extract
are not correct. We can still use the new hardware instructions by adding a
shuffle to move the elements to the correct indices. I introduced a new PPCISD
node here because adding a vector_shuffle and changing the elements of the
vector_extracts was getting undone by another DAG combine.
Commit on behalf of Zaara Syeda (syzaara@ca.ibm.com)
Differential Revision: https://reviews.llvm.org/D34009
llvm-svn: 307169
The existing check lines were more flexible, but these are
small enough tests that there shouldn't be much question
about register allocation. I've been hand-modifying this
file as I change the CGP memcmp expansion, but that's
more error-prone and time-consuming than just running the
update script.
llvm-svn: 306861
This patch fixes a verification error with -verify-machineinstrs while expanding __tls_get_addr by not creating ADJCALLSTACKUP and ADJCALLSTACKDOWN if there is another ADJCALLSTACKUP in this basic block since nesting ADJCALLSTACKUP/ADJCALLSTACKDOWN is not allowed.
Here, ADJCALLSTACKUP and ADJCALLSTACKDOWN are created as a fence for instruction scheduling to avoid _tls_get_addr is scheduled before mflr in the prologue (https://bugs.llvm.org//show_bug.cgi?id=25839). So if another ADJCALLSTACKUP exists before _tls_get_addr, we do not need to create a new ADJCALLSTACKUP.
Differential Revision: https://reviews.llvm.org/D34347
llvm-svn: 306678
As noted in D34071, there are some IR optimization opportunities that could be
handled by normal IR passes if this expansion wasn't happening so late in CGP.
Regardless of that, it seems wasteful to knowingly produce suboptimal IR here,
so I'm proposing this change:
%s = sub i32 %x, %y
%r = icmp ne %s, 0
=>
%r = icmp ne %x, %y
Changing the predicate to 'eq' mimics what InstCombine would do, so that's just
an efficiency improvement if we decide this expansion should happen sooner.
The fact that the PowerPC backend doesn't eliminate the 'subf.' might be
something for PPC folks to investigate separately.
Differential Revision: https://reviews.llvm.org/D34416
llvm-svn: 306471
When SelectionDAG merges consecutive stores and loads in MergeConsecutiveStores, it does not set dereferenceable flag for a created load instruction. This results in an assertion failure if SelectionDAG commonizes this load instruction with other load instructions, as well as it may miss optimization opportunities.
This patch sat dereferenceable flag for the newly created load instruction if all the load instructions to be merged are dereferenceable.
Differential Revision: https://reviews.llvm.org/D34679
llvm-svn: 306404
PowerPC backend does not pass the current optimization level to SelectionDAGISel and so SelectionDAGISel works with the default optimization level regardless of the current optimization level.
This patch makes the PowerPC backend set the optimization level correctly.
Differential Revision: https://reviews.llvm.org/D34615
llvm-svn: 306367
When SelectionDAG expands memcpy (or memmove) call into a sequence of load and store instructions, it disregards dereferenceable flag even the source pointer is known to be dereferenceable.
This results in an assertion failure if SelectionDAG commonizes a load instruction generated for memcpy with another load instruction for the source pointer.
This patch makes SelectionDAG to set the dereferenceable flag for the load instructions properly to avoid the assertion failure.
Differential Revision: https://reviews.llvm.org/D34467
llvm-svn: 306209
Define target hook isReallyTriviallyReMaterializable() to explicitly specify
PowerPC instructions that are trivially rematerializable. This will allow
the MachineLICM pass to accurately identify PPC instructions that should always
be hoisted.
Differential Revision: https://reviews.llvm.org/D34255
llvm-svn: 305932
This does some improvements/cleanup to the recently introduced
scavengeRegisterBackwards() functionality:
- Rewrite findSurvivorBackwards algorithm to use the existing
LiveRegUnit::accumulateBackward() code. This also avoids the Available
and Candidates bitset and just need 1 LiveRegUnit instance
(= 1 bitset).
- Pick registers in allocation order instead of register number order.
llvm-svn: 305817
This is the last step needed to avoid regressions for x86 before we flip the switch to allow
expansion of the smallest set of memcpy() via CGP. The DAG version checks for constant strings,
so we need to do that here too.
FWIW, the 2 constant test is not handled by LibCallSimplifier::optimizeMemCmp() because that
code is limited to 8-bit constant arrays. LibCallSimplifier will also fail to optimize some 1
constant tests because its alignment requirements are too strict (shouldn't require alignment
for a constant operand).
Differential Revision: https://reviews.llvm.org/D34071
llvm-svn: 305734
Re-apply r276044/r279124/r305516. Fixed a problem where we would refuse
to place spills as the very first instruciton of a basic block and thus
artifically increase pressure (test in
test/CodeGen/PowerPC/scavenging.mir:spill_at_begin)
This is a variant of scavengeRegister() that works for
enterBasicBlockEnd()/backward(). The benefit of the backward mode is
that it is not affected by incomplete kill flags.
This patch also changes
PrologEpilogInserter::doScavengeFrameVirtualRegs() to use the register
scavenger in backwards mode.
Differential Revision: http://reviews.llvm.org/D21885
llvm-svn: 305625
Revert because of reports of some PPC input starting to spill when it
was predicted that it wouldn't and no spillslot was reserved.
This reverts commit r305516.
llvm-svn: 305566
Re-apply r276044/r279124. Trying to reproduce or disprove the ppc64
problems reported in the stage2 build last time, which I cannot
reproduce right now.
This is a variant of scavengeRegister() that works for
enterBasicBlockEnd()/backward(). The benefit of the backward mode is
that it is not affected by incomplete kill flags.
This patch also changes
PrologEpilogInserter::doScavengeFrameVirtualRegs() to use the register
scavenger in backwards mode.
Differential Revision: http://reviews.llvm.org/D21885
llvm-svn: 305516
Add condition for MachineLICM to safely hoist instructions that utilize
non constant registers that are reserved.
On PPC, global variable access is done through the table of contents (TOC)
which is always in register X2. The ABI reserves this register in any
functions that have calls or access global variables.
A call through a function pointer involves saving, changing and restoring
this register around the call and thus MachineLICM does not consider it to
be invariant. We can however guarantee the register is preserved across the
call and thus is invariant.
Differential Revision: https://reviews.llvm.org/D33562
llvm-svn: 305490
This patch fixes a potential verification error (64-bit register operands for cmpw) with -verify-machineinstrs.
Differential Revision: https://reviews.llvm.org/D34208
llvm-svn: 305479
Power9 has instructions that will reverse the bytes within an element for all
sizes (half-word, word, double-word and quad-word). These can be used for the
vec_revb builtins in altivec.h. However, we implement these to match vector
shuffle nodes as that will cover both the builtins and vector shuffles that
occur in the SDAG through other means.
Differential Revision: https://reviews.llvm.org/D33690
llvm-svn: 305214
Note that if we need the result of both the divide and the modulo then we
compute the modulo based on the result of the divide and not using the new
hardware instruction.
Commit on behalf of STEFAN PINTILIE.
Differential Revision: https://reviews.llvm.org/D33940
llvm-svn: 305210
In PPCBoolRetToInt bool value is changed to i32 type. On ppc64 it may introduce an extra zero extension for the return value. This patch changes the integer type to i64 to avoid the zero extension on ppc64.
This patch fixed PR32442.
Differential Revision: https://reviews.llvm.org/D31407
llvm-svn: 305001
In SDAG, we don't expand libcalls with a nobuiltin attribute.
It's not clear if that's correct from the existing code comment:
"Don't do the check if marked as nobuiltin for some reason."
...adding a test here either way to show that there is currently
a different behavior implemented in the CGP-based expansion.
llvm-svn: 304991
The test diff for PowerPC shows we can better optimize if this case is one block.
For x86, there's would be a substantial difference if CGP expansion was enabled because branches are assumed
cheap and SDAG can't optimize across blocks.
Instead of this:
_cmp_eq8:
movq (%rdi), %rax
cmpq (%rsi), %rax
je LBB23_1
## BB#2: ## %res_block
movl $1, %ecx
jmp LBB23_3
LBB23_1:
xorl %ecx, %ecx
LBB23_3: ## %endblock
xorl %eax, %eax
testl %ecx, %ecx
sete %al
retq
We get this:
cmp_eq8:
movq (%rdi), %rcx
xorl %eax, %eax
cmpq (%rsi), %rcx
sete %al
retq
And that matches the optimal codegen that we get from the current expansion in SelectionDAGBuilder::visitMemCmpCall().
If this looks right, then I just need to confirm that vector-sized expansion will work from here, and we can enable
CGP memcmp() expansion for x86. Ie, we'll bypass the power-of-2 special cases currently optimized in SDAG because we
can lower the IR produced here optimally.
Differential Revision: https://reviews.llvm.org/D34005
llvm-svn: 304987
Lei Huang