This moves the X86 specific transform from rL364407
into DAGCombiner to generically handle 'little to big' cases
(for example: extract_subvector(v2i64 bitcast(v16i8))). This
allows us to remove both the x86 implementation and the aarch64
bitcast(extract_subvector(bitcast())) combine.
Earlier patches that dealt with regressions initially exposed
by this patch:
rG5e5e99c041e4
rG0b38af89e2c0
Patch by: @RKSimon (Simon Pilgrim)
Differential Revision: https://reviews.llvm.org/D63815
Summary:
The 2 source operands commutable instructions are encoded in the
VEX.VVVV field and the r/m field of the MODRM byte plus the VEX.B
field.
The VEX.B field is missing from the 2-byte VEX encoding. If the
VEX.VVVV source is 0-7 and the other register is 8-15 we can
swap them to avoid needing the VEX.B field. This works as long as
the VEX.W, VEX.mmmmm, and VEX.X fields are also not needed.
Fixes PR36706.
Reviewers: RKSimon, spatel
Reviewed By: RKSimon
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68550
gcc and icc pass these types in zmm registers in zmm registers.
This patch implements a quick hack to override the register
type before calling convention handling to one that is legal.
Longer term we might want to do something similar to 256-bit
integer registers on AVX1 where we just split all the operations.
Fixes PR42957
Differential Revision: https://reviews.llvm.org/D66708
llvm-svn: 370495
Summary: This is beneficial when the shuffle is only used once and end up being generated in a few places when some node is combined into a shuffle.
Reviewers: craig.topper, efriedma, RKSimon, lebedev.ri
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D66718
llvm-svn: 370326
This improves the combine I included in D66504 to handle constants in the upper operands of the concat. If we can constant fold them away we can pull the concat after the bin op. This helps with chains of madd reductions on X86 from loop unrolling. The loop madd reduction pattern creates pmaddwd with half the width of the add that follows it using zeroes to fill the upper bits. If we have two of these added together we can pull the zeroes through the accumulating add and then shrink it.
Differential Revision: https://reviews.llvm.org/D66680
llvm-svn: 369937
The AVX2 check lines show two issues. An ADD that became an OR
because we knew the input was disjoint, but really it was zero
so we should have just removed the ADD/OR all together.
Relatedly we use 128-bit VPMADDWD instructions followed by
256-bit VPADDD operations. We should be able to narrow these
VPADDDs.
llvm-svn: 369736
The assert that caused this to be reverted should be fixed now.
Original commit message:
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 368183
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 367901
The pmaddwd inserts a truncate, if that truncate would end up
creating additional instructions instead of making a zext
narrower, then we shouldn't do it.
I've restricted this to only sse4.1 targets since on prior
targets the zext will be done in stages. So the truncate will
probably not create additional instructions. Might need some
more investigation of mul shrinking and the other pmaddwd
transform to be sure this is the right decision.
There might be a slight regression on AVX1 targets due to add
splitting. Hard to say for sure. Maybe we need to look into
using the vector reduction flag to use 2 narrow loads and a
blend instead of extracting and inserting.
llvm-svn: 367198
This reverts r340478 and r340631 and replaces them with a simpler
method of just letting DAG combine revisit the nodes to handle
the other operand.
llvm-svn: 367195
I noticed that we were failing to narrow an x86 ymm math op in a case similar
to the 'madd' test diff. That is because a bitcast is sitting between the math
and the extract subvector and thwarting our pattern matching for narrowing:
t56: v8i32 = add t59, t58
t68: v4i64 = bitcast t56
t73: v2i64 = extract_subvector t68, Constant:i64<2>
t96: v4i32 = bitcast t73
There are a few wins and neutral diffs in the other tests.
Differential Revision: https://reviews.llvm.org/D61806
llvm-svn: 360541
This is the sibling fold for insert-of-insert that was added with D56604.
Now that we have x86 shuffle narrowing (D57156), this change shows improvements for
lots of AVX512 reduction code (not sure that we would ever expect extract-of-extract otherwise).
There's a small regression in some of the partial-permute tests (extracting followed by splat).
That is tracked by PR40500:
https://bugs.llvm.org/show_bug.cgi?id=40500
Differential Revision: https://reviews.llvm.org/D57336
llvm-svn: 352528
This isn't the final fix for our reduction/horizontal codegen, but it takes care
of a lot of the problems. After we narrow the shuffle, existing combines for
insert/extract and binops kick in, and we end up with cheaper 128-bit ops.
The avg and mul reduction tests show an existing shuffle lowering hole for
AVX2/AVX512. I think in its most minimal form this is:
https://bugs.llvm.org/show_bug.cgi?id=40434
...but we might need multiple fixes to get it right.
Differential Revision: https://reviews.llvm.org/D57156
llvm-svn: 352209
This seems to be getting in the way more than its helping. This does mean we stop scalarizing some cases, but I'm not convinced the scalarization was really better.
Some of the changes to vsel-cmp-load.ll are a regression but D56156 should fix it.
llvm-svn: 350159
Summary:
Make machine PHIs optimization to work for single value register taken from
several different copies. This is the first step to fix PR38917. This change
allows to get rid of redundant PHIs (see opt_phis2.mir test) to make
the subsequent optimizations (like CSE) possible and simpler.
For instance, before this patch the code like this:
%b = COPY %z
...
%a = PHI %bb1, %a; %bb2, %b
could be optimized to:
%a = %b
but the code like this:
%c = COPY %z
...
%b = COPY %z
...
%a = PHI %bb1, %a; %bb2, %b; %bb3, %c
would remain unchanged.
With this patch the latter case will be optimized:
%a = %z```.
Committed on behalf of: Anton Afanasyev anton.a.afanasyev@gmail.com
Reviewers: RKSimon, MatzeB
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54839
llvm-svn: 349271
Previously if V2 was unused we ended up using V1 for both inputs as part of the code that follows the new code. By using lowerVectorShuffleWithUNPCK we keep the undef nature of V2 in the output.
As near as I can tell this makes v16i8 behavior consistent with every other VT now.
This does mean that we give the register allocator freedom to fill in random registers now and create false dependencies. But like I said we're already doing that for other types.
llvm-svn: 347296
This is a long-awaited follow-up suggested in D33578. Since then, we've picked up even more
opportunities for vector narrowing from changes like D53784, so there are a lot of test diffs.
Apart from 2-3 strange cases, these are all wins.
I've structured this to be no-functional-change-intended for any target except for x86
because I couldn't tell if AArch64, ARM, and AMDGPU would improve or not. All of those
targets have existing regression tests (4, 4, 10 files respectively) that would be
affected. Also, Hexagon overrides the shouldReduceLoadWidth() hook, but doesn't show
any regression test diffs. The trade-off is deciding if an extra vector load is better
than a single wide load + extract_subvector.
For x86, this is almost always better (on paper at least) because we often can fold
loads into subsequent ops and not increase the official instruction count. There's also
some unknown -- but potentially large -- benefit from using narrower vector ops if wide
ops are implemented with multiple uops and/or frequency throttling is avoided.
Differential Revision: https://reviews.llvm.org/D54073
llvm-svn: 346595
The majority of the changes are because the rest of shuffle lowering/combining prefers to replace the undef input with the other operand. Using UNPCKL directly seemed to avoid this and just grabbed a randomish register for the undef which can create false dependencies.
llvm-svn: 346050
We already have custom lowering for the AVX case in LegalizeVectorOps. So its better to keep the regular extend op around as long as possible.
I had to qualify one place in DAG combine that created illegal vector extending load operations. This change by itself had no effect on any tests which is why its included here.
I've made a few cleanups to the custom lowering. The sign extend code no longer creates an identity shuffle with undef elements. The zero extend code now emits a zero_extend_vector_inreg instead of an unpckl with a zero vector.
For the high half of the custom lowering of zero_extend/any_extend, we're now using an unpckh with a zero vector or undef. Previously we used used a pshufd to move the upper 64-bits to the lower 64-bits and then used a zero_extend_vector_inreg. I think the zero vector should require less execution resources and be smaller code size.
Differential Revision: https://reviews.llvm.org/D54024
llvm-svn: 346043
Narrowing vector binops came up in the demanded bits discussion in D52912.
I don't think we're going to be able to do this transform in IR as a canonicalization
because of the risk of creating unsupported widths for vector ops, but we already have
a DAG TLI hook to allow what I was hoping for: isExtractSubvectorCheap(). This is
currently enabled for x86, ARM, and AArch64 (although only x86 has existing regression
test diffs).
This is artificially limited to not look through bitcasts because there are so many
test diffs already, but that's marked with a TODO and is a small follow-up.
Differential Revision: https://reviews.llvm.org/D53784
llvm-svn: 345602
This patch brings back the MOV64r0 pseudo instruction for zeroing a 64-bit register. This replaces the SUBREG_TO_REG MOV32r0 sequence we use today. Post register allocation we will rewrite the MOV64r0 to a 32-bit xor with an implicit def of the 64-bit register similar to what we do for the various XMM/YMM/ZMM zeroing pseudos.
My main motivation is to enable the spill optimization in foldMemoryOperandImpl. As we were seeing some code that repeatedly did "xor eax, eax; store eax;" to spill several registers with a new xor for each store. With this optimization enabled we get a store of a 0 immediate instead of an xor. Though I admit the ideal solution would be one xor where there are multiple spills. I don't believe we have a test case that shows this optimization in here. I'll see if I can try to reduce one from the code were looking at.
There's definitely some other machine CSE(and maybe other passes) behavior changes exposed by this patch. So it seems like there might be some other deficiencies in SUBREG_TO_REG handling.
Differential Revision: https://reviews.llvm.org/D52757
llvm-svn: 345165
This is the planned follow-up to D52997. Here we are reducing horizontal vector math codegen
by default. AMD Jaguar (btver2) should have no difference with this patch because it has
fast-hops. (If we want to set that bit for other CPUs, let me know.)
The code changes are small, but there are many test diffs. For files that are specifically
testing for hops, I added RUNs to distinguish fast/slow, so we can see the consequences
side-by-side. For files that are primarily concerned with codegen other than hops, I just
updated the CHECK lines to reflect the new default codegen.
To recap the recent horizontal op story:
1. Before rL343727, we were producing hops for all subtargets for a variety of patterns.
Hops were likely not optimal for all targets though.
2. The IR improvement in r343727 exposed a hole in the backend hop pattern matching, so
we reduced hop codegen for all subtargets. That was bad for Jaguar (PR39195).
3. We restored the hop codegen for all targets with rL344141. Good for Jaguar, but
probably bad for other CPUs.
4. This patch allows us to distinguish when we want to produce hops, so everyone can be
happy. I'm not sure if we have the best predicate here, but the intent is to undo the
extra hop-iness that was enabled by r344141.
Differential Revision: https://reviews.llvm.org/D53095
llvm-svn: 344361
Enable enableMultipleCopyHints() on X86.
Original Patch by @jonpa:
While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling.
Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates.
Differential Revision: https://reviews.llvm.org/D38128
llvm-svn: 342578
As discussed on D41794, we have many cases where we fail to combine shuffles as the input operands have other uses.
This patch permits these shuffles to be combined as long as they don't introduce additional variable shuffle masks, which should reduce instruction dependencies and allow the total number of shuffles to still drop without increasing the constant pool.
However, this may mean that some memory folds may no longer occur, and on pre-AVX require the occasional extra register move.
This also exposes some poor PMULDQ/PMULUDQ codegen which was doing unnecessary upper/lower calculations which will in fact fold to zero/undef - the fix will be added in a followup commit.
Differential Revision: https://reviews.llvm.org/D50328
llvm-svn: 339335
Summary:
This is the pattern you get from the loop vectorizer for something like this
int16_t A[1024];
int16_t B[1024];
int32_t C[512];
void pmaddwd() {
for (int i = 0; i != 512; ++i)
C[i] = (A[2*i]*B[2*i]) + (A[2*i+1]*B[2*i+1]);
}
In this case we will have (add (mul (build_vector), (build_vector)), (mul (build_vector), (build_vector))). This is different than the pattern we currently match which has the build_vectors between an add and a single multiply. I'm not sure what C code would get you that pattern.
Reviewers: RKSimon, spatel, zvi
Reviewed By: zvi
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D49636
llvm-svn: 338097
This seems to be a net improvement. There's still an issue under avx512f where we have a 512-bit vpaddd, but not vpmaddwd so we end up doing two 256-bit vpmaddwds and inserting the results before a 512-bit vpaddd. It might be better to do two 512-bits paddds with zeros in the upper half. Same number of instructions, but breaks a dependency.
llvm-svn: 337656
In DAGCombiner, we try to simplify this pattern:
([s|z]ext (load ...))
Conceptually, a new extload which is created while splitting the load
should have the same debug location as the load.
Making this change affects the IROrder of the new load, causing some
test case churn.
In practice, the new location is never different from the location of
the [s|z]ext, at least not during check-llvm or a stage2 build.
Part of: llvm.org/PR37262
Differential Revision: https://reviews.llvm.org/D46156
llvm-svn: 331301
In addition to the existing match as part of a loop-reduction, add a
straightforward pattern match for DAG-contained patterns.
Reviewers: RKSimon, craig.topper
Subscribers: llvm-commits
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D41811
llvm-svn: 322446
As part of the unification of the debug format and the MIR format, print
MBB references as '%bb.5'.
The MIR printer prints the IR name of a MBB only for block definitions.
* find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g'
* find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g'
* find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g'
* grep -nr 'BB#' and fix
Differential Revision: https://reviews.llvm.org/D40422
llvm-svn: 319665
Currently if we're inserting 0s into the upper elements of a vector register we insert an explicit move of the smaller register to implicitly zero the upper bits. But if we can prove that they are already zero we can skip that. This is based on a similar idea of what we do to avoid emitting explicit zero extends for GR32->GR64.
Unfortunately, this is harder for vector registers because there are several opcodes that don't have VEX equivalent instructions, but can write to XMM registers. Among these are SHA instructions and a MMX->XMM move. Bitcasts can also get in the way.
So for now I'm starting with explicitly allowing only VPMADDWD because we emit zeros in combineLoopMAddPattern. So that is placing extra instruction into the reduction loop.
I'd like to allow PSADBW as well after D37453, but that's currently blocked by a bitcast. We either need to peek through bitcasts or canonicalize insert_subvectors with zeros to remove bitcasts on the value being inserted.
Longer term we should probably have a cleanup pass that removes superfluous zeroing moves even when the producer is in another basic block which is something these isel tricks can't do. See PR32544.
Differential Revision: https://reviews.llvm.org/D37653
llvm-svn: 313365
Ideally we'd be able to emit the SUBREG_TO_REG without the explicit register->register move, but we'd need to be sure the producing operation would select something that guaranteed the upper bits were already zeroed.
llvm-svn: 312450
In a future patch, I plan to teach isel to use a small vector move with implicit zeroing of the upper elements when it sees the (insert_subvector zero, X, 0) pattern.
llvm-svn: 312448
Summary:
This teaches 512-bit shuffles to detect unused halfs in order to reduce shuffle size.
We may need to refine the 512-bit exit point. I couldn't remember if we had good cross lane shuffles for 8/16 bit with AVX-512 or not.
I believe this is step towards being able to handle D36454 without a special case.
From here we need to improve our ability to combine extract_subvector with insert_subvector and other extract_subvectors. And we need to support narrowing binary operations where we don't demand all elements. This may be improvements to DAGCombiner::narrowExtractedVectorBinOp(by recognizing an insert_subvector in addition to concat) or we may need a target specific combiner.
Reviewers: RKSimon, zvi, delena, jbhateja
Reviewed By: RKSimon, jbhateja
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D36601
llvm-svn: 310724
The root cause of reverting was fixed - PR33514.
Summary:
The patch makes instruction count the highest priority for
LSR solution for X86 (previously registers had highest priority).
Reviewers: qcolombet
Differential Revision: http://reviews.llvm.org/D30562
From: Evgeny Stupachenko <evstupac@gmail.com>
<evgeny.v.stupachenko@intel.com>
llvm-svn: 310289
This seems to be interacting badly with ASan somehow, causing false reports of
heap-buffer overflows: PR33514.
> Summary:
> The patch makes instruction count the highest priority for
> LSR solution for X86 (previously registers had highest priority).
>
> Reviewers: qcolombet
>
> Differential Revision: http://reviews.llvm.org/D30562
>
> From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 305720
Summary:
The patch makes instruction count the highest priority for
LSR solution for X86 (previously registers had highest priority).
Reviewers: qcolombet
Differential Revision: http://reviews.llvm.org/D30562
From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 304824
Sanjay Patel