One motivation for making this change is that the lack of using movmsk is likely
a main source of perf difference between clang and gcc on the C-Ray benchmark as
shown here:
https://www.phoronix.com/scan.php?page=article&item=gcc-clang-2019&num=5
...but this change alone isn't enough to solve that problem.
The 'all-of' examples show what is likely the worst case trade-off: we end up with
an extra instruction (or 2 if we count the 'xor' register clearing). The 'any-of'
examples look clearly better using movmsk because we've traded 2 vector instructions
for 2 scalar instructions, and movmsk may have better timing than the generic 'movq'.
If we examine the llvm-mca output for these cases, it appears that even though the
'all-of' movmsk variant looks worse on paper, it would perform better on both
Haswell and Jaguar.
$ llvm-mca -mcpu=haswell no_movmsk.s -timeline
Iterations: 100
Instructions: 400
Total Cycles: 504
Total uOps: 400
Dispatch Width: 4
uOps Per Cycle: 0.79
IPC: 0.79
Block RThroughput: 1.0
$ llvm-mca -mcpu=haswell movmsk.s -timeline
Iterations: 100
Instructions: 600
Total Cycles: 358
Total uOps: 600
Dispatch Width: 4
uOps Per Cycle: 1.68
IPC: 1.68
Block RThroughput: 1.5
$ llvm-mca -mcpu=btver2 no_movmsk.s -timeline
Iterations: 100
Instructions: 400
Total Cycles: 407
Total uOps: 400
Dispatch Width: 2
uOps Per Cycle: 0.98
IPC: 0.98
Block RThroughput: 2.0
$ llvm-mca -mcpu=btver2 movmsk.s -timeline
Iterations: 100
Instructions: 600
Total Cycles: 311
Total uOps: 600
Dispatch Width: 2
uOps Per Cycle: 1.93
IPC: 1.93
Block RThroughput: 3.0
Finally, there may be CPUs where movmsk is horribly slow (old AMD small cores?), but if
that's true, then we're also almost certainly making the wrong transform already for
reductions with >2 elements, so that should be fixed independently.
Differential Revision: https://reviews.llvm.org/D59997
llvm-svn: 357367
Fixes PR41316 where the expanded PAVG intrinsic had had one of its ADDs turned into an OR due to its operands having no conflicting bits.
llvm-svn: 357351
We need XMM registers to handle varargs with the Win64 ABI. Before we would
silently generate bad code resulting in an assertion failure elsewhere in the
backend.
llvm-svn: 357317
This is probably the least important of our movmsk problems, but I'm starting
at the bottom to reduce distractions.
We were creating a select_cc which bypasses the select and bitmask codegen
optimizations that we have now. If we produce a compare+negate instead, we
allow things like neg/sbb carry bit hacks, and in all cases we avoid a cmov.
There's no partial register update danger in these sequences because we always
produce the zero-register xor ahead of the 'set' if needed.
There seems to be a missing fold for sext of a bool bit here:
negl %ecx
movslq %ecx, %rax
...but that's an independent transform.
Differential Revision: https://reviews.llvm.org/D59818
llvm-svn: 357172
If we know the 2 halves of an oversized zext-in-reg are the same,
don't create those halves independently.
I tried several different approaches to fold this, but it's difficult
to get right during legalization. In the default path, we are creating
a generic shuffle that looks like an unpack high, but it can get
transformed into a different mask (a blend), so it's not
straightforward to match that. If we try to fold after it actually
becomes an X86ISD::UNPCKH node, we can't be sure what the operand node
is - it might be a generic shuffle, or it could be some x86-specific op.
From the test output, we should be doing something like this for SSE4.1
as well, but I'd rather leave that as a follow-up since it involves
changing lowering actions.
Differential Revision: https://reviews.llvm.org/D59777
llvm-svn: 357129
This is not exactly NFC because it should make further combines
of MOVMSK easier to match, but there should be no outward differences
because we have isel patterns in place specifically to allow this. See:
// Also support integer VTs to avoid a int->fp bitcast in the DAG.
llvm-svn: 357128
Enable SSE41 ZERO_EXTEND_VECTOR_INREG shuffle combines - for the PMOVZX(PSHUFD(V)) -> UNPCKH(V,0) pattern we reduce the shuffles (port5-bottleneck on Intel) at the expense of creating a zero (pxor v,v) and an extra register move - which is a good trade off as these are pretty cheap and in most cases it doesn't increase register pressure.
This also exposed a missed opportunity to use combine to ZERO_EXTEND_VECTOR_INREG with folded loads - even if we're in the float domain.
........
Causes PR41249
llvm-svn: 357057
Enable SSE41 ZERO_EXTEND_VECTOR_INREG shuffle combines - for the PMOVZX(PSHUFD(V)) -> UNPCKH(V,0) pattern we reduce the shuffles (port5-bottleneck on Intel) at the expense of creating a zero (pxor v,v) and an extra register move - which is a good trade off as these are pretty cheap and in most cases it doesn't increase register pressure.
This also exposed a missed opportunity to use combine to ZERO_EXTEND_VECTOR_INREG with folded loads - even if we're in the float domain.
llvm-svn: 356864
Just enable this for AVX for now as SSE41 introduces extra register moves for the PMOVZX(PSHUFD(V)) -> UNPCKH(V,0) pattern (but otherwise helps reduce port5 usage on Intel targets).
Only AVX support is required for PR40685 as the issue is due to 8i8->8i32 zext shuffle leftovers.
llvm-svn: 356858
This is yet another step towards solving PR14613:
https://bugs.llvm.org/show_bug.cgi?id=14613
uaddsat X, Y --> (X >u (X + Y)) ? -1 : X + Y
usubsat X, Y --> (X >u Y) ? X - Y : 0
We can't count on a sane vector ISA, so override the default (umin/umax)
expansion of unsigned add/sub saturate in cases where we do not have umin/umax.
Differential Revision: https://reviews.llvm.org/D59006
llvm-svn: 356855
On 32-bit targets without popcnt, we currently expand 64-bit popcnt to sequences of arithmetic and logic ops for each 32-bit half and then add the 32 bit halves together. If we have xmm registers we can use use those to implement the operation instead. This results in less instructions then doing two separate 32-bit popcnt sequences.
This mitigates some of PR41151 for the i64 on i686 case when we have SSE2.
Differential Revision: https://reviews.llvm.org/D59662
llvm-svn: 356808
We used a lock cmpxchg8b to do i64 atomic loads. But if we have SSE2 we can do better and use a plain movq to do the load instead.
I tried to just use an f64 atomic load and add isel patterns to MOVSD(which the domain fixing pass can turn to MOVQ), but the atomic_load SDNode in TargetSelectionDAG.td requires the type to be integer.
So I've emitted VZEXT_LOAD instead which should be selected by isel to a MOVQ. Hopefully we don't need a specific atomic flavor of this. I kept the memory operand from the original AtomicSDNode. I wasn't sure if I might need to set the MOVolatile flag?
I've left some FIXMEs for improvements we can do without SSE2.
Differential Revision: https://reviews.llvm.org/D59679
llvm-svn: 356807
CMPXCHG8B was introduced on i586/pentium generation.
If its not enabled, limit the atomic width to 32 bits so the AtomicExpandPass will expand to lib calls. Unclear if we should be using a different limit for other configs. The default is 1024 and experimentation shows that using an i256 atomic will cause a crash in SelectionDAG.
Differential Revision: https://reviews.llvm.org/D59576
llvm-svn: 356631
This patch enables the use of lowerShuffleAsBitMask for 512-bit blends before
falling back to move immedate, GPR to k-register, and masked op.
I had to make some changes to support v8i64 when i64 is not a legal type. And to
support floating point types.
This trades a load for the move immediate and GPR move which is higher latency.
But its probably better for register pressure not having to hop through other
register classes. The load+and should play better with LICM and
rematerialization I think.
Differential Revision: https://reviews.llvm.org/D59479
llvm-svn: 356618
This patch removes the following dag node opcodes from namespace X86ISD:
RDTSC_DAG,
RDTSCP_DAG,
RDPMC_DAG
The logic that expands RDTSC/RDPMC/XGETBV intrinsics is basically the same. The
only differences are:
RDTSC/RDTSCP don't implicitly read ECX.
RDTSCP also implicitly writes ECX.
I moved the common expansion logic into a helper function with the goal to get
rid of code repetition. That helper is now used for the expansion of
RDTSC/RDTSCP/RDPMC/XGETBV intrinsics.
No functional change intended.
Differential Revision: https://reviews.llvm.org/D59547
llvm-svn: 356546
These changes are related to PR37743 and include:
SelectionDAGBuilder::visitSelect handles the unary SelectPatternFlavor::SPF_ABS case to build ABS node.
Delete the redundant recognizer of the integer ABS pattern from the DAGCombiner.
Add promoting the integer ABS node in the LegalizeIntegerType.
Expand-based legalization of integer result for the ABS nodes.
Expand-based legalization of ABS vector operations.
Add some integer abs testcases for different typesizes for Thumb arch
Add the custom ABS expanding and change the SAD pattern recognizer for X86 arch: The i64 result of the ABS is expanded to:
tmp = (SRA, Hi, 31)
Lo = (UADDO tmp, Lo)
Hi = (XOR tmp, (ADDCARRY tmp, hi, Lo:1))
Lo = (XOR tmp, Lo)
The "detectZextAbsDiff" function is changed for the recognition of pattern with the ABS node. Given a ABS node, detect the following pattern:
(ABS (SUB (ZERO_EXTEND a), (ZERO_EXTEND b))).
Change integer abs testcases for codegen with the ABS node support for AArch64.
Indicate that the ABS is legal for the i64 type when the NEON is supported.
Change the integer abs testcases to show changing of codegen.
Add combine and legalization of ABS nodes for Thumb arch.
Extend 'matchSelectPattern' to recognize the ABS patterns with ICMP_SGE condition.
For discussion, see https://bugs.llvm.org/show_bug.cgi?id=37743
Patch by: @ikulagin (Ivan Kulagin)
Differential Revision: https://reviews.llvm.org/D49837
llvm-svn: 356468
This allows better code size for aarch64 floating point materialization
in a future patch.
Reviewers: evandro
Differential Revision: https://reviews.llvm.org/D58690
llvm-svn: 356389
Summary:
As noted by @andreadb in https://reviews.llvm.org/D59035#inline-525780
If we have `sext (trunc (cmov C0, C1) to i8)`,
we can instead do `cmov (sext (trunc C0 to i8)), (sext (trunc C1 to i8))`
Reviewers: craig.topper, andreadb, RKSimon
Reviewed By: craig.topper
Subscribers: llvm-commits, andreadb
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59412
llvm-svn: 356301
The asm parser generates the immediate without the SAE bit. So for consistency we should generate the MCInst the same way from CodeGen.
Since they are now both the same, remove the masking from the printer and replace with an llvm_unreachable.
Use a target constant since we're rebuilding the node anyway. Then we don't have to have isel convert it. Saves about 500 bytes from the isel table.
llvm-svn: 356294
Reduce the size of an any-extended i64 scalar_to_vector source to i32 - the any_extend nodes are often introduced by SimplifyDemandedBits.
llvm-svn: 356292
This is an immediate fix for:
https://bugs.llvm.org/show_bug.cgi?id=41066
...but as noted there and the code comments, we should do better
by stubbing this out sooner.
llvm-svn: 356158
The feature flag alone can't be trusted since it can be passed via -mattr. Need to ensure 64-bit mode as well.
We had a 64 bit mode check on the instruction to make the assembler work correctly. But we weren't guarding any of our lowering code or the hooks for the AtomicExpandPass.
I've added 32-bit command lines to atomic128.ll with and without cx16. The tests there would all previously fail if -mattr=cx16 was passed to them. I had to move one test case for f128 to a new file as it seems to have a different 32-bit mode or possibly sse issue.
Differential Revision: https://reviews.llvm.org/D59308
llvm-svn: 356078
Attempt to combine CONCAT_VECTORS nodes, which we only really have pre-legalization.
This encourages a lot of X86ISD::SUBV_BROADCAST generation, so I've added SimplifyDemandedVectorEltsForTargetNode handling for this at the same time.
The X86ISD::VTRUNC regression in shuffle-vs-trunc-256-widen.ll will be handled in a future commit.
llvm-svn: 356064
A fuzzer found the crasher:
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=13700
The bug was introduced recently here:
rL355741
This is the quick fix. If we need to do this transform
later, then we'd have to extend/truncate the vector setcc
element type to the scalar setcc type (i8).
llvm-svn: 356053
AVX1 broadcasts were failing as we were adding bitcasts that caused MayFoldLoad's hasOneUse to return false.
This patch stops introducing bitcasts so early and also replaces the broadcast index scaling through bitcasts (which can't succeed in some cases) to instead just keep track of the bitoffset which can be converted back to the broadcast index later on.
Differential Revision: https://reviews.llvm.org/D58888
llvm-svn: 356043
Krzysztof Parzyszek