This change enables soft-float for PowerPC64, and also makes soft-float disable
all vector instruction sets for both 32-bit and 64-bit modes. This latter part
is necessary because the PPC backend canonicalizes many Altivec vector types to
floating-point types, and so soft-float breaks scalarization support for many
operations. Both for embedded targets and for operating-system kernels desiring
soft-float support, it seems reasonable that disabling hardware floating-point
also disables vector instructions (embedded targets without hardware floating
point support are unlikely to have Altivec, etc. and operating system kernels
desiring not to use floating-point registers to lower syscall cost are unlikely
to want to use vector registers either). If someone needs this to work, we'll
need to change the fact that we promote many Altivec operations to act on
v4f32. To make it possible to disable Altivec when soft-float is enabled,
hardware floating-point support needs to be expressed as a positive feature,
like the others, and not a negative feature, because target features cannot
have dependencies on the disabling of some other feature. So +soft-float has
now become -hard-float.
Fixes PR26970.
llvm-svn: 283060
Now we can commute to BLENDPD/BLENDPS on SSE41+ targets if necessary, so simplify the combine matching where we can.
This required me to add a couple of scalar math movsd/moss fold patterns that hadn't been needed in the past.
llvm-svn: 283038
Instead of selecting between MOVSD/MOVSS and BLENDPD/BLENDPS at shuffle lowering by subtarget this will help us select the instruction based on actual commutation requirements.
We could possibly add BLENDPD/BLENDPS -> MOVSD/MOVSS commutation and MOVSD/MOVSS memory folding using a similar approach if it proves useful
I avoided adding AVX512 handling as I'm not sure when we should be making use of VBLENDPD/VBLENDPS on EVEX targets
llvm-svn: 283037
-Remove OptForSize. Not all of the backend follows the same rules for creating broadcasts and there is no conflicting pattern.
-Don't stop selecting VEX VMOVDDUP when AVX512 is supported. We need VLX for EVEX VMOVDDUP.
-Only use VMOVDDUP for v2i64 broadcasts if AVX2 is not supported.
llvm-svn: 283020
We can't use Jcc to leave a Win64 function in general, because that
confuses the unwinder. However, for "leaf" functions, that is, functions
where the return address is always on top of the stack and which don't
have unwind info, it's OK.
Differential Revision: https://reviews.llvm.org/D24836
llvm-svn: 282920
Register stackification currently checks VNInfo for changes. Make that
more accurate by testing each intervening instruction for any other defs
to the same virtual register.
Patch by Jacob Gravelle
Differential Revision: https://reviews.llvm.org/D24942
llvm-svn: 282886
Summary:
Previously, when allocating unspillable live ranges, we would never
attempt to split. We would always bail out and try last ditch graph
recoloring.
This patch changes this by attempting to split all live intervals before
performing recoloring.
This fixes LLVM bug PR14879.
I can't add test cases for any backends other than AVR because none of
them have small enough register classes to trigger the bug.
Reviewers: qcolombet
Subscribers: MatzeB
Differential Revision: https://reviews.llvm.org/D25070
llvm-svn: 282852
If AVX512 is disabled, the registers should already be marked reserved. Pattern predicates and register classes on instructions should take care of most of the rest. Loads/stores and physical register copies for XMM16-31 and YMM16-31 without VLX have already been taken care of.
I'm a little unclear why this changed the register allocation of the SSE2 run of the sad.ll test, but the registers selected appear to be valid after this change.
llvm-svn: 282835
For some reason there are both of these available, except
for scalar 64-bit compares which only has u64. I'm not sure
why there are both (I'm guessing it's for the one bit inputs we
don't use), but for consistency always using the
unsigned one.
llvm-svn: 282832
This adds new pseudo instructions that can be selected during register allocation to represent loads and stores of XMM/YMM registers when AVX512F is available, but VLX isn't. They will be converted to VEX encoded moves if the register turns out to be XMM0-15/YMM0-15. Otherwise either an EVEX VEXTRACT(store) or VBROADCAST(load) will be used.
Fixes one of the cases from PR29112.
llvm-svn: 282690
Fixes to allow spilling all registers at the end of the block
work with exec modifications. Don't emit s_and_saveexec_b64 for
if lowering, and instead emit copies. Mark control flow mask
instructions as terminators to get correct spill code placement
with fast regalloc, and then have a separate optimization pass
form the saveexec.
This should work if SGPRs are spilled to VGPRs, but
will likely fail in the case that an SGPR spills to memory
and no workitem takes a divergent branch.
llvm-svn: 282667
Normally, if conversion would add implicit uses for redefined registers,
e.g. R0<def> = add_if ..., R0<imp-use>. However, if only subregisters of
R0 are known to be live but not R0 itself, such implicit uses will not be
added, causing prior definitions of such subregisters and R0 itself to
become dead.
llvm-svn: 282626
This addresses PR26055 LiveDebugValues is very slow.
Contrary to the old LiveDebugVariables pass LiveDebugValues currently
doesn't look at the lexical scopes before inserting a DBG_VALUE
intrinsic. This means that we often propagate DBG_VALUEs much further
down than necessary. This is especially noticeable in large C++
functions with many inlined method calls that all use the same
"this"-pointer.
For example, in the following code it makes no sense to propagate the
inlined variable a from the first inlined call to f() into any of the
subsequent basic blocks, because the variable will always be out of
scope:
void sink(int a);
void __attribute((always_inline)) f(int a) { sink(a); }
void foo(int i) {
f(i);
if (i)
f(i);
f(i);
}
This patch reuses the LexicalScopes infrastructure we have for
LiveDebugVariables to take this into account.
The effect on compile time and memory consumption is quite noticeable:
I tested a benchmark that is a large C++ source with an enormous
amount of inlined "this"-pointers that would previously eat >24GiB
(most of them for DBG_VALUE intrinsics) and whose compile time was
dominated by LiveDebugValues. With this patch applied the memory
consumption is 1GiB and 1.7% of the time is spent in LiveDebugValues.
https://reviews.llvm.org/D24994
Thanks to Daniel Berlin and Keith Walker for reviewing!
llvm-svn: 282611
Simplify Consecutive Merge Store Candidate Search
Now that address aliasing is much less conservative, push through
simplified store merging search which only checks for parallel stores
through the chain subgraph. This is cleaner as the separation of
non-interfering loads/stores from the store-merging logic.
Whem merging stores, search up the chain through a single load, and
finds all possible stores by looking down from through a load and a
TokenFactor to all stores visited. This improves the quality of the
output SelectionDAG and generally the output CodeGen (with some
exceptions).
Additional Minor Changes:
1. Finishes removing unused AliasLoad code
2. Unifies the the chain aggregation in the merged stores across
code paths
3. Re-add the Store node to the worklist after calling
SimplifyDemandedBits.
4. Increase GatherAllAliasesMaxDepth from 6 to 18. That number is
arbitrary, but seemed sufficient to not cause regressions in
tests.
This finishes the change Matt Arsenault started in r246307 and
jyknight's original patch.
Many tests required some changes as memory operations are now
reorderable. Some tests relying on the order were changed to use
volatile memory operations
Noteworthy tests:
CodeGen/AArch64/argument-blocks.ll -
It's not entirely clear what the test_varargs_stackalign test is
supposed to be asserting, but the new code looks right.
CodeGen/AArch64/arm64-memset-inline.lli -
CodeGen/AArch64/arm64-stur.ll -
CodeGen/ARM/memset-inline.ll -
The backend now generates *worse* code due to store merging
succeeding, as we do do a 16-byte constant-zero store efficiently.
CodeGen/AArch64/merge-store.ll -
Improved, but there still seems to be an extraneous vector insert
from an element to itself?
CodeGen/PowerPC/ppc64-align-long-double.ll -
Worse code emitted in this case, due to the improved store->load
forwarding.
CodeGen/X86/dag-merge-fast-accesses.ll -
CodeGen/X86/MergeConsecutiveStores.ll -
CodeGen/X86/stores-merging.ll -
CodeGen/Mips/load-store-left-right.ll -
Restored correct merging of non-aligned stores
CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll -
Improved. Correctly merges buffer_store_dword calls
CodeGen/AMDGPU/si-triv-disjoint-mem-access.ll -
Improved. Sidesteps loading a stored value and merges two stores
CodeGen/X86/pr18023.ll -
This test has been removed, as it was asserting incorrect
behavior. Non-volatile stores *CAN* be moved past volatile loads,
and now are.
CodeGen/X86/vector-idiv.ll -
CodeGen/X86/vector-lzcnt-128.ll -
It's basically impossible to tell what these tests are actually
testing. But, looks like the code got better due to the memory
operations being recognized as non-aliasing.
CodeGen/X86/win32-eh.ll -
Both loads of the securitycookie are now merged.
CodeGen/AMDGPU/vgpr-spill-emergency-stack-slot-compute.ll -
This test appears to work but no longer exhibits the spill
behavior.
Reviewers: arsenm, hfinkel, tstellarAMD, nhaehnle, jyknight
Subscribers: wdng, nhaehnle, nemanjai, arsenm, weimingz, niravd, RKSimon, aemerson, qcolombet, resistor, tstellarAMD, t.p.northover, spatel
Differential Revision: https://reviews.llvm.org/D14834
llvm-svn: 282600
The KORTEST was introduced due to a bug where a TEST instruction used a K register.
but, turns out that the opposite case of KORTEST using a GPR is now happening
The change removes the KORTEST flow and adds a COPY instruction from the K reg to a GPR.
Differential Revision: https://reviews.llvm.org/D24953
llvm-svn: 282580
This check currently doesn't seem to do anything useful on any in-tree target:
On non-x86, it always evaluates to false, so we never hit the code path that
creates the shuffle with zero.
On x86, it just forwards to isShuffleMaskLegal(), which is a reasonable thing to
query in general, but doesn't make sense if only restricted to zero blends.
Differential Revision: https://reviews.llvm.org/D24625
llvm-svn: 282567
The 'or' case shows up in copysign. The copysign code also had
redundant checking for a scalar zero operand with 'and', so I
removed that.
I'm not sure how to test vector 'and', 'andn', and 'xor' yet,
but it seems better to just include all of the logic ops since
we're fixing 'or' anyway.
llvm-svn: 282546
Summary:
The current implementation of isConstantPhysReg() checks for defs of
physical registers to determine if they are constant. Some
architectures (e.g. AArch64 XZR/WZR) have registers that are constant
and may be used as destinations to indicate the generated value is
discarded, preventing isConstantPhysReg() from returning true. This
change adds a TargetRegisterInfo hook that overrides the no defs check
for cases such as this.
Reviewers: MatzeB, qcolombet, t.p.northover, jmolloy
Subscribers: junbuml, aemerson, mcrosier, rengolin
Differential Revision: https://reviews.llvm.org/D24570
llvm-svn: 282543
Variables are sometimes missing their debug location information in
blocks in which the variables should be available. This would occur
when one or more predecessor blocks had not yet been visited by the
routine which propagated the information from predecessor blocks.
This is addressed by only considering predecessor blocks which have
already been visited.
The solution to this problem was suggested by Daniel Berlin on the
LLVM developer mailing list.
Differential Revision: https://reviews.llvm.org/D24927
llvm-svn: 282506
Disable tail calls while the remaining bugs are fixed. Enable only for tests.
Reviewers: vkalintiris
Differential Review: https://reviews.llvm.org/D24912
llvm-svn: 282487
This patch corresponds to review:
https://reviews.llvm.org/D24396
This patch adds support for the "vector count trailing zeroes",
"vector compare not equal" and "vector compare not equal or zero instructions"
as well as "scalar count trailing zeroes" instructions. It also changes the
vector negation to use XXLNOR (when VSX is enabled) so as not to increase
register pressure (previously this was done with a splat immediate of all
ones followed by an XXLXOR). This was done because the altivec.h
builtins (patch to follow) use vector negation and the use of an additional
register for the splat immediate is not optimal.
llvm-svn: 282478
When we have dynamic allocas we have a frame pointer, and
when we're lowering frame indexes we should make sure we use it.
Patch by Jacob Gravelle
Differential Revision: https://reviews.llvm.org/D24889
llvm-svn: 282442
Don't match the UXTW extended reg forms of ADD/ADDS/SUB/SUBS if the
32-bit to 64-bit zero-extend can be done for free by taking advantage
of the 32-bit defining instruction zeroing the upper 32-bits of the X
register destination. This enables better instruction selection in a
few cases, such as:
sub x0, xzr, x8
instead of:
mov x8, xzr
sub x0, x8, w9, uxtw
madd x0, x1, x1, x8
instead of:
mul x9, x1, x1
add x0, x9, w8, uxtw
cmp x2, x8
instead of:
sub x8, x2, w8, uxtw
cmp x8, #0
add x0, x8, x1, lsl #3
instead of:
lsl x9, x1, #3
add x0, x9, w8, uxtw
Reviewers: t.p.northover, jmolloy
Subscribers: mcrosier, aemerson, llvm-commits, rengolin
Differential Revision: https://reviews.llvm.org/D24747
llvm-svn: 282413
