It turns out that SchedWrite WriteIMulH was always assigned to the low half of
the result of a MULX (rather than to the high half).
To avoid confusion, this patch swaps the two MULX writes in the tablegen
definition of MULX32/64. That way, write names better describe what they
actually refer to; this also avoids further complications if in future we decide
to reuse the same MulH writes to also model other scalar integer multiply
instructions. I also had to swap the latency values for the two MULX writes to
make sure that the change is effectively an NFC. In fact, none of the existing
x86 tests were affected by this small refactoring.
This patch also fixes a bug in MCA: a wrong latency value was propagated for
instructions that perform multiple writes to a same register. This last issue
was found by Roman while testing MULX on targets that define a different latency
for the Low/High part of the result.
Differential Revision: https://reviews.llvm.org/D108727
Before this patch, WriteIMulH reported a latency value which is correct for the
RR variant of MULX, but not for the RM variant.
This patch fixes the issue by introducing a new WriteIMulHLd, which is meant to
be used only by the RM variant of MULX.
Differential Revision: https://reviews.llvm.org/D108701
Before this patch, instructions MULX32rm and MULX64rm were missing a ReadAdvance
for the implicit read of register EDX/RDX. This patch fixes the issue, and it
also introduces a new SchedWrite for the two variants of MULX. The general idea
behind this last change is to eventually decrease the number of InstRW in the
scheduling models.
This patch also adds a ReadAdvance for the implicit read of EFLAGS in ADCX/ADOX.
Differential Revision: https://reviews.llvm.org/D108372
Much like `mulx`'s `WriteIMulH`, there are two outputs of
AVX2 GATHER instructions. This was changed back in rL160110,
but the sched model change wasn't present.
So right now, for sched models that are marked as complete
(`znver3` only now), codegen'ning `GATHER` results in a crash:
```
DefIdx 1 exceeds machine model writes for early-clobber renamable $ymm3, dead early-clobber renamable $ymm2 = VPGATHERDDYrm killed renamable $ymm3(tied-def 0), undef renamable $rax, 4, renamable $ymm0, 0, $noreg, killed renamable $ymm2(tied-def 1) :: (load 32, align 1)
```
https://godbolt.org/z/Ks7zW7WGh
I'm guessing we need to deal with this like we deal with `WriteIMulH`.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D104205
We never codegen them so this doesn't matter in practice. But
sometimes someone comes along and tries to use these flags
for something else. LIke the Load Value Inject inline assembly
handling.
This effectively splits the scheduling WriteVecMaskedStore(Y) classes
into four different classes (one per each variant).
The new VecMaskedStore scheduling classes are now correctly marked as
'unsupported' by the bdver2 and btver2 models.
No functional change intended.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D80201
uops.info says these should be 15 cycle instructions. Uops.info also shows the 512-bit form uses port 0 and 5 for both register and memory. We had memory using 0 and 1.
Differential Revision: https://reviews.llvm.org/D75549
Based on uops.info these should have 5 cycle latency as they did on Haswell/Broadwell. I have no additional internal information from Intel.
This was also shown as a discrepancy in the spreadsheet that was sent with an early llvm-dev post about llvm-exegesis.
It also matches Agner Fog.
Differential Revision: https://reviews.llvm.org/D74357
The load ports need a cycle for each potentially loaded element just like Haswell and Skylake. Unlike Haswell and Broadwell, the number of uops does not scale with the number of elements. Instead the load uops run for multiple cycles.
I've taken the latency number from the uops.info. The port binding for the non-load uops is taken from the original IACA data I have.
Differential Revision: https://reviews.llvm.org/D74000
Most X87 compare instructions write to the X87 status word, while the SSE (U)COMI compares write to rFLAGS. These are often handled very differently on CPUs (e.g. rFLAGS outputs typically involve a fpu2gpr transfer), and we shouldn't be grouping all these instructions behind a single class - so this patch splits off the SSE compares into a new WriteFComX class (and currently keeps the same behaviours). If there's a need to distinguish between X87 instructions more closely we can investigate that in the future, but as we don't handle any of the X87 side effects at the moment its unlikely to have any notable effect.
On BtVer2 conditional SIMD stores are heavily microcoded.
The latency is directly proportional to the number of packed elements extracted
from the input vector. Also, according to micro-benchmarks, most of the
computation seems to be done in the integer unit.
Only a minority of the uOPs is executed by the FPU. The observed behaviour on
the FPU looks similar to this:
- The input MASK value is moved to the Integer Unit
-- [ a VMOVMSK-like uOP-executed on JFPU0].
- In parallel, each element of the input XMM/YMM is extracted and then sent to
the IntegerUnit through JFPU1.
As expected, a (conditional) store is executed for every extracted element.
Interestingly, a (speculative) load is executed for every extracted element too.
It is as-if a "LOAD - BIT_EXTRACT- CMOV" sequence of uOPs is repeated by the
integer unit for every contionally stored element.
VMASKMOVDQU is a special case: the number of speculative loads is always 2
(presumably, one load per quadword). That means, extra shifts and masking is
performed on (one of) the loaded quadwords before each conditional store (that
also explains the big number of non-FP uOPs retired).
This patch replaces the existing writes for conditional SIMD stores (i.e.
WriteFMaskedStore, and WriteFMaskedStoreY) with the following new writes:
WriteFMaskedStore32 [ XMM Packed Single ]
WriteFMaskedStore32Y [ YMM Packed Single ]
WriteFMaskedStore64 [ XMM Packed Double ]
WriteFMaskedStore64Y [ YMM Packed Double ]
Added a wrapper class named X86SchedWriteMaskMove in X86Schedule.td to describe
both RM and MR variants for conditional SIMD moves in a single tablegen
definition.
Instances of that class are then passed in input to multiclass avx_movmask_rm
when constructing MASKMOVPS/PD definitions.
Since this patch introduces new writes, I had to update all the X86 scheduling
models.
Differential Revision: https://reviews.llvm.org/D66801
llvm-svn: 370649
This copies the Sandy Bridge zero idiom support to later CPUs. Adding the AVX2 and AVX512F/VL instructions as appropriate.
Differential Revision: https://reviews.llvm.org/D62360
llvm-svn: 361690
Summary:
This avoids needing an isel pattern for each condition code. And it removes translation switches for converting between SETcc instructions and condition codes.
Now the printer, encoder and disassembler take care of converting the immediate. We use InstAliases to handle the assembly matching. But we print using the asm string in the instruction definition. The instruction itself is marked IsCodeGenOnly=1 to hide it from the assembly parser.
Reviewers: andreadb, courbet, RKSimon, spatel, lebedev.ri
Reviewed By: andreadb
Subscribers: hiraditya, lebedev.ri, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60138
llvm-svn: 357801
Summary:
Reorder the condition code enum to match their encodings. Move it to MC layer so it can be used by the scheduler models.
This avoids needing an isel pattern for each condition code. And it removes
translation switches for converting between CMOV instructions and condition
codes.
Now the printer, encoder and disassembler take care of converting the immediate.
We use InstAliases to handle the assembly matching. But we print using the
asm string in the instruction definition. The instruction itself is marked
IsCodeGenOnly=1 to hide it from the assembly parser.
This does complicate the scheduler models a little since we can't assign the
A and BE instructions to a separate class now.
I plan to make similar changes for SETcc and Jcc.
Reviewers: RKSimon, spatel, lebedev.ri, andreadb, courbet
Reviewed By: RKSimon
Subscribers: gchatelet, hiraditya, kristina, lebedev.ri, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D60041
llvm-svn: 357800
Rotate with explicit immediate is a single uop from Haswell on. An immediate of 1 has a dependency on the previous writer of flags, but the other immediate values do not.
The implicit rotate by 1 instruction is 2 uops. But the flags are merged after the rotate uop so the data result does not see the flag dependency. But I don't think we have any way of modeling that.
RORX is 1 uop without the load. 2 uops with the load. We currently model these with WriteShift/WriteShiftLd.
Differential Revision: https://reviews.llvm.org/D59077
llvm-svn: 355636
Summary:
The AX/EAX/RAX with immediate forms are 2 uops just like the AL with immediate.
The modrm form with r8 and immediate is a single uop just like r16/r32/r64 with immediate.
Reviewers: RKSimon, andreadb
Reviewed By: RKSimon
Subscribers: gbedwell, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58581
llvm-svn: 354754
This patch adds a new ReadAdvance definition named ReadInt2Fpu.
ReadInt2Fpu allows x86 scheduling models to accurately describe delays caused by
data transfers from the integer unit to the floating point unit.
ReadInt2Fpu currently defaults to a delay of zero cycles (i.e. no delay) for all
x86 models excluding BtVer2. That means, this patch is only a functional change
for the Jaguar cpu model only.
Tablegen definitions for instructions (V)PINSR* have been updated to account for
the new ReadInt2Fpu. That read is mapped to the the GPR input operand.
On Jaguar, int-to-fpu transfers are modeled as a +6cy delay. Before this patch,
that extra delay was added to the opcode latency. In practice, the insert opcode
only executes for 1cy. Most of the actual latency is actually contributed by the
so-called operand-latency. According to the AMD SOG for family 16h, (V)PINSR*
latency is defined by expression f+1, where f is defined as a forwarding delay
from the integer unit to the fpu.
When printing instruction latency from MCA (see InstructionInfoView.cpp) and LLC
(only when flag -print-schedule is speified), we now need to account for any
extra forwarding delays. We do this by checking if scheduling classes declare
any negative ReadAdvance entries. Quoting a code comment in TargetSchedule.td:
"A negative advance effectively increases latency, which may be used for
cross-domain stalls". When computing the instruction latency for the purpose of
our scheduling tests, we now add any extra delay to the formula. This avoids
regressing existing codegen and mca schedule tests. It comes with the cost of an
extra (but very simple) hook in MCSchedModel.
Differential Revision: https://reviews.llvm.org/D57056
llvm-svn: 351965
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
Currently we hardcode instructions with ReadAfterLd if the register operands don't need to be available until the folded load has completed. This doesn't take into account the different load latencies of different memory operands (PR36957).
This patch adds a ReadAfterFold def into X86FoldableSchedWrite to replace ReadAfterLd, allowing us to specify the load latency at a scheduler class level.
I've added ReadAfterVec*Ld classes that match the XMM/Scl, XMM and YMM/ZMM WriteVecLoad classes that we currently use, we can tweak these values in future patches once this infrastructure is in place.
Differential Revision: https://reviews.llvm.org/D52886
llvm-svn: 343868
Summary:
While looking at PR35606, I found out that the scheduling info is incorrect.
One can check that it's really a P5+P6 and not a 2*P56 with:
echo -e 'vzeroall\nvandps %xmm1, %xmm2, %xmm3' | ./bin/llvm-exegesis -mode=uops -snippets-file=-
(vandps executes on P5 only)
Reviewers: craig.topper, RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D52541
llvm-svn: 343447
As suggested by Craig Topper - I'm going to look at cleaning up the RMW sequences instead.
The uops are slightly different to the register variant, so requires a +1uop tweak
llvm-svn: 342969
We're missing quite a bit of data for these instruction, removing the overrides makes this obvious - inconsistent reg/mem variants is a concern as well.
Also, we have Divider resources (HWDivider etc.) but they aren't actually used consistently.
llvm-svn: 342904
Split WriteIMul by size and also by IMUL multiply-by-imm and multiply-by-reg cases.
This removes all the scheduler overrides for gpr multiplies and stops WriteMULH being ignored for BMI2 MULX instructions.
llvm-svn: 342892
Variable Shifts/Rotates using the CL register have different behaviours to the immediate instructions - split accordingly to help remove yet more repeated overrides from the schedule models.
llvm-svn: 342852
Confirmed with Craig Topper - fix a typo that was missing a Port4 uop for ROR*mCL instructions on some Intel models.
Yet another step on the scheduler model cleanup marathon......
llvm-svn: 342846
NFCI for now, but it should make it easier to remove a lot of unnecessary overrides in a future commit.
Now that funnel shift intrinsics are coming online we need to get this cleaned up to make vectorization costs from scalar rotate patterns more straightforward.
llvm-svn: 342837
Don't declare them as X86SchedWritePair when the folded class will never be used.
Note: MOVBE (load/store endian conversion) instructions tend to have a very different behaviour to BSWAP.
llvm-svn: 338412
Andrea Di Biagio