| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch introduces the following changes to the btver2 scheduling model:
- The number of micro opcodes for YMM loads and stores is now 2 (it was
incorrectly set to 1 for both aligned and misaligned loads/stores).
- Increased the number of AGU resource cycles for YMM loads and stores
to 2cy (instead of 1cy).
- Removed JFPU01 and JFPX from the list of resources consumed by pure
float/vector loads (no MMX).
I verified with llvm-exegesis that pure XMM/YMM loads are no-pipe. Those
are dispatched to the FPU but not really issues on JFPU01.
Differential Revision: https://reviews.llvm.org/D68871
llvm-svn: 374765
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
As disscused in https://bugs.llvm.org/show_bug.cgi?id=43219,
i believe it may be somewhat useful to show //some// aggregates
over all the sea of statistics provided.
Example:
```
Average Wait times (based on the timeline view):
[0]: Executions
[1]: Average time spent waiting in a scheduler's queue
[2]: Average time spent waiting in a scheduler's queue while ready
[3]: Average time elapsed from WB until retire stage
[0] [1] [2] [3]
0. 3 1.0 1.0 4.7 vmulps %xmm0, %xmm1, %xmm2
1. 3 2.7 0.0 2.3 vhaddps %xmm2, %xmm2, %xmm3
2. 3 6.0 0.0 0.0 vhaddps %xmm3, %xmm3, %xmm4
3 3.2 0.3 2.3 <total>
```
I.e. we average the averages.
Reviewers: andreadb, mattd, RKSimon
Reviewed By: andreadb
Subscribers: gbedwell, arphaman, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68714
llvm-svn: 374361
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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
|
|
|
|
|
|
|
|
|
|
|
|
| |
CMPXCHG8B/16B.
This is a follow up of r369642.
This patch assigns a ReadAfterLd to every implicit register use of instruction
CMPXCHG8B and instruction CMPXCHG16B. Perf micro-benchmarks show that implicit
registers are read after 3cy from the start of execution.
llvm-svn: 369750
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Excluding ADC/SBB and the bit-test instructions (BTR/BTS/BTC), the observed
latency of all other RMW integer arithmetic/logic instructions is 6cy and not
5cy.
Example (ADD):
```
addb $0, (%rsp) # Latency: 6cy
addb $7, (%rsp) # Latency: 6cy
addb %sil, (%rsp) # Latency: 6cy
addw $0, (%rsp) # Latency: 6cy
addw $511, (%rsp) # Latency: 6cy
addw %si, (%rsp) # Latency: 6cy
addl $0, (%rsp) # Latency: 6cy
addl $511, (%rsp) # Latency: 6cy
addl %esi, (%rsp) # Latency: 6cy
addq $0, (%rsp) # Latency: 6cy
addq $511, (%rsp) # Latency: 6cy
addq %rsi, (%rsp) # Latency: 6cy
```
The same latency profile applies to SUB/AND/OR/XOR/INC/DEC.
The observed latency of ADC/SBB is 7-8cy. So we need a different write to model
those. Latency of BTS/BTR/BTC is not fixed by this patch (they are much slower
than what the model for btver2 currently reports).
Differential Revision: https://reviews.llvm.org/D66636
llvm-svn: 369748
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Single operand MUL instructions that implicitly set EAX have the following
latency/throughput profile (see below):
imul %cl # latency: 3cy - uOPs: 1 - 1 JMul
imul %cx # latency: 3cy - uOPs: 3 - 3 JMul
imul %ecx # latency: 3cy - uOPs: 2 - 2 JMul
imul %rcx # latency: 6cy - uOPs: 2 - 4 JMul
mul %cl # latency: 3cy - uOPs: 1 - 1 JMul
mul %cx # latency: 3cy - uOPs: 3 - 3 JMul
mul %ecx # latency: 3cy - uOPs: 2 - 2 JMul
mul %rcx # latency: 6cy - uOPs: 2 - 4 JMul
Excluding the 64bit variant, which has a latency of 6cy, every other instruction
has a latency of 3cy. However, the number of decoded macro-opcodes (as well as
the resource cyles) depend on the MUL size.
The two operand MULs have a more predictable profile (see below):
imul %dx, %dx # latency: 3cy - uOPs: 1 - 1 JMul
imul %edx, %edx # latency: 3cy - uOPs: 1 - 1 JMul
imul %rdx, %rdx # latency: 6cy - uOPs: 1 - 4 JMul
imul $3, %dx, %dx # latency: 4cy - uOPs: 2 - 2 JMul
imul $3, %ecx, %ecx # latency: 3cy - uOPs: 1 - 1 JMul
imul $3, %rdx, %rdx # latency: 6cy - uOPs: 1 - 4 JMul
This patch updates the values in the Jaguar scheduling model and regenerates
llvm-mca tests.
Differential Revision: https://reviews.llvm.org/D66547
llvm-svn: 369661
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On Jaguar, XCHG has a latency of 1cy and decodes to 2 macro-opcodes. Maximum
throughput for XCHG is 1 IPC. The byte exchange has worse latency and decodes to
1 extra uOP; maximum observed throughput is 0.5 IPC.
```
xchgb %cl, %dl # Latency: 2cy - uOPs: 3 - 2 ALU
xchgw %cx, %dx # Latency: 1cy - uOPs: 2 - 2 ALU
xchgl %ecx, %edx # Latency: 1cy - uOPs: 2 - 2 ALU
xchgq %rcx, %rdx # Latency: 1cy - uOPs: 2 - 2 ALU
```
The reg-mem forms of XCHG are atomic operations with an observed latency of
16cy. The resource usage is similar to the XCHGrr variants. The biggest
difference is obviously the bus-locking, which prevents the LS to issue other
memory uOPs in parallel until the unlocking store uOP is executed.
```
xchgb %cl, (%rsp) # Latency: 16cy - uOPs: 3 - ECX latency: 11cy
xchgw %cx, (%rsp) # Latency: 16cy - uOPs: 3 - ECX latency: 11cy
xchgl %ecx, (%rsp) # Latency: 16cy - uOPs: 3 - ECX latency: 11cy
xchgq %rcx, (%rsp) # Latency: 16cy - uOPs: 3 - ECX latency: 11cy
```
The exchanged in/out register operand becomes available after 11cy from the
start of execution. Added test xchg.s to verify that we correctly see that
register write committed in 11cy (and not 16cy).
Reg-reg XADD instructions have the same latency/throughput than the byte
exchange (register-register variant).
```
xaddb %cl, %dl # latency: 2cy - uOPs: 3 - 3 ALU
xaddw %cx, %dx # latency: 2cy - uOPs: 3 - 3 ALU
xaddl %ecx, %edx # latency: 2cy - uOPs: 3 - 3 ALU
xaddq %rcx, %rdx # latency: 2cy - uOPs: 3 - 3 ALU
```
The non-atomic RM variants have a latency of 11cy, and decode to 4
macro-opcodes. They still consume 2 ALU pipes, and the exchange in/out register
operand becomes available in 3cy (it matches the 'load-to-use latency').
```
xaddb %cl, (%rsp) # latency: 11cy - uOPs: 4 - 3 ALU
xaddw %cx, (%rsp) # latency: 11cy - uOPs: 4 - 3 ALU
xaddl %ecx, (%rsp) # latency: 11cy - uOPs: 4 - 3 ALU
xaddq %rcx, (%rsp) # latency: 11cy - uOPs: 4 - 3 ALU
```
The atomic XADD variants execute in 16cy. The in/out register operand is
available after 11cy from the start of execution.
```
lock xaddb %cl, (%rsp) # latency: 16cy - uOPs: 4 - 3 ALU -- ECX latency: 11cy
lock xaddw %cx, (%rsp) # latency: 16cy - uOPs: 4 - 3 ALU -- ECX latency: 11cy
lock xaddl %ecx, (%rsp) # latency: 16cy - uOPs: 4 - 3 ALU -- ECX latency: 11cy
lock xaddq %rcx, (%rsp) # latency: 16cy - uOPs: 4 - 3 ALU -- ECX latency: 11cy
```
Added test xadd.s to verify those latencies as well as read-advance values.
Differential Revision: https://reviews.llvm.org/D66535
llvm-svn: 369642
|
|
|
|
|
|
| |
This is a follow-up of r369365.
llvm-svn: 369412
|
|
|
|
|
|
|
|
|
| |
Latency and throughput of LOCK INC/DEC/NEG/NOT is always 19cy.
Number of uOPs is still 1.
Differential Revision: https://reviews.llvm.org/D66469
llvm-svn: 369388
|
|
|
|
|
|
| |
D66424 adds the base support for LOCK so we should be able to add special case support for all these cases in future patches
llvm-svn: 369367
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On Jaguar, CMPXCHG has a latency of 11cy, and a maximum throughput of 0.33 IPC.
Throughput is superiorly limited to 0.33 because of the implicit in/out
dependency on register EAX. In the case of repeated non-atomic CMPXCHG with the
same memory location, store-to-load forwarding occurs and values for sequent
loads are quickly forwarded from the store buffer.
Interestingly, the functionality in LLVM that computes the reciprocal throughput
doesn't seem to know about RMW instructions. That functionality only looks at
the "consumed resource cycles" for the throughput computation. It should be
fixed/improved by a future patch. In particular, for RMW instructions, that
logic should also take into account for the write latency of in/out register
operands.
An atomic CMPXCHG has a latency of ~17cy. Throughput is also limited to
~17cy/inst due to cache locking, which prevents other memory uOPs to start
executing before the "lock releasing" store uOP.
CMPXCHG8rr and CMPXCHG8rm are treated specially because they decode to one less
macro opcode. Their latency tend to be the same as the other RR/RM variants. RR
variants are relatively fast 3cy (but still microcoded - 5 macro opcodes).
CMPXCHG8B is 11cy and unfortunately doesn't seem to benefit from store-to-load
forwarding. That means, throughput is clearly limited by the in/out dependency
on GPR registers. The uOP composition is sadly unknown (due to the lack of PMCs
for the Integer pipes). I have reused the same mix of consumed resource from the
other CMPXCHG instructions for CMPXCHG8B too.
LOCK CMPXCHG8B is instead 18cycles.
CMPXCHG16B is 32cycles. Up to 38cycles when the LOCK prefix is specified. Due to
the in/out dependencies, throughput is limited to 1 instruction every 32 (or 38)
cycles dependeing on whether the LOCK prefix is specified or not.
I wouldn't be surprised if the microcode for CMPXCHG16B is similar to 2x
microcode from CMPXCHG8B. So, I have speculatively set the JALU01 consumption to
2x the resource cycles used for CMPXCHG8B.
The two new hasLockPrefix() functions are used by the btver2 scheduling model
check if a MCInst/MachineInst has a LOCK prefix. Calls to hasLockPrefix() have
been encoded in predicates of variant scheduling classes that describe lat/thr
of CMPXCHG.
Differential Revision: https://reviews.llvm.org/D66424
llvm-svn: 369365
|
|
|
|
|
|
|
|
|
|
| |
resources-x86_64.s files. NFC
In D66424 it has been requested to move all the new tests added by r369278 into
resources-x86_64.s. That is because only the 8b/16 ops should be tested by
resources-cmpxchg.s. This partially reverts r369278.
llvm-svn: 369288
|
|
|
|
|
|
| |
Addresses a review comment in D66424
llvm-svn: 369279
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
instructions based on the simulation.
This patch teaches the bottleneck analysis how to identify and print the most
expensive sequence of instructions according to the simulation. Fixes PR37494.
The goal is to help users identify the sequence of instruction which is most
critical for performance.
A dependency graph is internally used by the bottleneck analysis to describe
data dependencies and processor resource interferences between instructions.
There is one node in the graph for every instruction in the input assembly
sequence. The number of nodes in the graph is independent from the number of
iterations simulated by the tool. It means that a single node of the graph
represents all the possible instances of a same instruction contributed by the
simulated iterations.
Edges are dynamically "discovered" by the bottleneck analysis by observing
instruction state transitions and "backend pressure increase" events generated
by the Execute stage. Information from the events is used to identify critical
dependencies, and materialize edges in the graph. A dependency edge is uniquely
identified by a pair of node identifiers plus an instance of struct
DependencyEdge::Dependency (which provides more details about the actual
dependency kind).
The bottleneck analysis internally ranks dependency edges based on their impact
on the runtime (see field DependencyEdge::Dependency::Cost). To this end, each
edge of the graph has an associated cost. By default, the cost of an edge is a
function of its latency (in cycles). In practice, the cost of an edge is also a
function of the number of cycles where the dependency has been seen as
'contributing to backend pressure increases'. The idea is that the higher the
cost of an edge, the higher is the impact of the dependency on performance. To
put it in another way, the cost of an edge is a measure of criticality for
performance.
Note how a same edge may be found in multiple iteration of the simulated loop.
The logic that adds new edges to the graph checks if an equivalent dependency
already exists (duplicate edges are not allowed). If an equivalent dependency
edge is found, field DependencyEdge::Frequency of that edge is incremented by
one, and the new cost is cumulatively added to the existing edge cost.
At the end of simulation, costs are propagated to nodes through the edges of the
graph. The goal is to identify a critical sequence from a node of the root-set
(composed by node of the graph with no predecessors) to a 'sink node' with no
successors. Note that the graph is intentionally kept acyclic to minimize the
complexity of the critical sequence computation algorithm (complexity is
currently linear in the number of nodes in the graph).
The critical path is finally computed as a sequence of dependency edges. For
edges describing processor resource interferences, the view also prints a
so-called "interference probability" value (by dividing field
DependencyEdge::Frequency by the total number of iterations).
Examples of critical sequence computations can be found in tests added/modified
by this patch.
On output streams that support colored output, instructions from the critical
sequence are rendered with a different color.
Strictly speaking the analysis conducted by the bottleneck analysis view is not
a critical path analysis. The cost of an edge doesn't only depend on the
dependency latency. More importantly, the cost of a same edge may be computed
differently by different iterations.
The number of dependencies is discovered dynamically based on the events
generated by the simulator. However, their number is not fixed. This is
especially true for edges that model processor resource interferences; an
interference may not occur in every iteration. For that reason, it makes sense
to also print out a "probability of interference".
By construction, the accuracy of this analysis (as always) is strongly dependent
on the simulation (and therefore the quality of the information available in the
scheduling model).
That being said, the critical sequence effectively identifies a performance
criticality. Instructions from that sequence are expected to have a very big
impact on performance. So, users can take advantage of this information to focus
their attention on specific interactions between instructions.
In my experience, it works quite well in practice, and produces useful
output (in a reasonable amount time).
Differential Revision: https://reviews.llvm.org/D63543
llvm-svn: 364045
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
llvm.x86.sse.stmxcsr only writes to memory.
llvm.x86.sse.ldmxcsr only reads from memory, and might generate an FPE.
Reviewers: craig.topper, RKSimon
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62896
llvm-svn: 363773
|
|
|
|
|
|
|
| |
Bottleneck Analysis is one of the many views available in llvm-mca. Therefore,
it should be enabled when flag -all-views is passed in input to the tool.
llvm-svn: 362964
|
|
|
|
|
|
|
|
| |
This fixes a problem where back-pressure increases caused by register
dependencies were not correctly notified if execution was also delayed by memory
dependencies.
llvm-svn: 361740
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
printing.
We require d/q suffixes on the memory form of these instructions to disambiguate the memory size.
We don't require it on the register forms, but need to support parsing both with and without it.
Previously we always printed the d/q suffix on the register forms, but it's redundant and
inconsistent with gcc and objdump.
After this patch we should support the d/q for parsing, but not print it when its unneeded.
llvm-svn: 360085
|
|
|
|
|
|
| |
This is defined as part of SSE1, XMM PMOVMSKB doesn't appear until SSE2
llvm-svn: 359477
|
|
|
|
| |
llvm-svn: 358160
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
parser and disassembly parser to remove inconsistencies between VEX and EVEX.
Many of our instructions have both a _Int form used by intrinsics and a form
used by other IR constructs. In the EVEX space the _Int versions usually cover
all the capabilities include broadcasting and rounding. While the other version
only covers simple register/register or register/load forms. For this reason
in EVEX, the non intrinsic form is usually marked isCodeGenOnly=1.
In the VEX encoding space we were less consistent, but usually the _Int version
was the isCodeGenOnly version.
This commit makes the VEX instructions match the EVEX instructions. This was
done by manually studying the AsmMatcher table so its possible I missed some
cases, but we should be closer now.
I'm thinking about using the isCodeGenOnly bit to simplify the EVEX2VEX
tablegen code that disambiguates the _Int and non _Int versions. Currently it
checks register class sizes and Record the memory operands come from. I have
some other changes I was looking into for D59266 that may break the memory check.
I had to make a few scheduler hacks to keep the _Int versions from being treated
differently than the non _Int version.
Differential Revision: https://reviews.llvm.org/D60441
llvm-svn: 358138
|
|
|
|
|
|
|
|
|
| |
It makes more sense to print out the number of micro opcodes that are issued
every cycle rather than the number of instructions issued per cycle.
This behavior is also consistent with the dispatch-stats: numbers from the two
views can now be easily compared.
llvm-svn: 357919
|
|
|
|
|
|
|
|
|
|
| |
custom printing and custom parsing to achieve the same result and more
Similar to previous change done for VPCOM and VPCMP
Differential Revision: https://reviews.llvm.org/D59468
llvm-svn: 356384
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
Since bottleneck hints are enabled via user request, it can be
confusing if no bottleneck information is presented. Such is the
case when no bottlenecks are identified. This patch emits a message
in that case.
Reviewers: andreadb
Reviewed By: andreadb
Subscribers: tschuett, gbedwell, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D59098
llvm-svn: 355628
|
|
|
|
|
|
| |
Some targets have fast-path handling for these patterns that we should model.
llvm-svn: 355498
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch adds a new flag named -bottleneck-analysis to print out information
about throughput bottlenecks.
MCA knows how to identify and classify dynamic dispatch stalls. However, it
doesn't know how to analyze and highlight kernel bottlenecks. The goal of this
patch is to teach MCA how to correlate increases in backend pressure to backend
stalls (and therefore, the loss of throughput).
From a Scheduler point of view, backend pressure is a function of the scheduler
buffer usage (i.e. how the number of uOps in the scheduler buffers changes over
time). Backend pressure increases (or decreases) when there is a mismatch
between the number of opcodes dispatched, and the number of opcodes issued in
the same cycle. Since buffer resources are limited, continuous increases in
backend pressure would eventually leads to dispatch stalls. So, there is a
strong correlation between dispatch stalls, and how backpressure changed over
time.
This patch teaches how to identify situations where backend pressure increases
due to:
- unavailable pipeline resources.
- data dependencies.
Data dependencies may delay execution of instructions and therefore increase the
time that uOps have to spend in the scheduler buffers. That often translates to
an increase in backend pressure which may eventually lead to a bottleneck.
Contention on pipeline resources may also delay execution of instructions, and
lead to a temporary increase in backend pressure.
Internally, the Scheduler classifies instructions based on whether register /
memory operands are available or not.
An instruction is marked as "ready to execute" only if data dependencies are
fully resolved.
Every cycle, the Scheduler attempts to execute all instructions that are ready
to execute. If an instruction cannot execute because of unavailable pipeline
resources, then the Scheduler internally updates a BusyResourceUnits mask with
the ID of each unavailable resource.
ExecuteStage is responsible for tracking changes in backend pressure. If backend
pressure increases during a cycle because of contention on pipeline resources,
then ExecuteStage sends a "backend pressure" event to the listeners.
That event would contain information about instructions delayed by resource
pressure, as well as the BusyResourceUnits mask.
Note that ExecuteStage also knows how to identify situations where backpressure
increased because of delays introduced by data dependencies.
The SummaryView observes "backend pressure" events and prints out a "bottleneck
report".
Example of bottleneck report:
```
Cycles with backend pressure increase [ 99.89% ]
Throughput Bottlenecks:
Resource Pressure [ 0.00% ]
Data Dependencies: [ 99.89% ]
- Register Dependencies [ 0.00% ]
- Memory Dependencies [ 99.89% ]
```
A bottleneck report is printed out only if increases in backend pressure
eventually caused backend stalls.
About the time complexity:
Time complexity is linear in the number of instructions in the
Scheduler::PendingSet.
The average slowdown tends to be in the range of ~5-6%.
For memory intensive kernels, the slowdown can be significant if flag
-noalias=false is specified. In the worst case scenario I have observed a
slowdown of ~30% when flag -noalias=false was specified.
We can definitely recover part of that slowdown if we optimize class LSUnit (by
doing extra bookkeeping to speedup queries). For now, this new analysis is
disabled by default, and it can be enabled via flag -bottleneck-analysis. Users
of MCA as a library can enable the generation of pressure events through the
constructor of ExecuteStage.
This patch partially addresses https://bugs.llvm.org/show_bug.cgi?id=37494
Differential Revision: https://reviews.llvm.org/D58728
llvm-svn: 355308
|
|
|
|
|
|
|
|
|
|
|
| |
dispatch stalls.
Dispatch stall cycles may be associated to multiple dispatch stall events.
Before this patch, each stall cycle was associated with a single stall event.
This patch also improves a couple of code comments, and adds a helper method to
query the Scheduler for dispatch stalls.
llvm-svn: 354877
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes a bug where register writes performed by optimizable register
moves were sometimes wrongly treated like partial register updates. Before this
patch, llvm-mca wrongly predicted a 1.50 IPC for test reg-move-elimination-6.s
(added by this patch). With this patch, llvm-mca correctly updates the register
defintions in the PRF, and the IPC for that test is now correctly reported as 2.
llvm-svn: 354271
|
|
|
|
|
|
|
|
|
|
| |
arguments where on is %st.
All of these instructions consume one encoded register and the other register is %st. They either write the result to %st or the encoded register. Previously we printed both arguments when the encoded register was written. And we printed one argument when the result was written to %st. For the stack popping forms the encoded register is always the destination and we didn't print both operands. This was inconsistent with gcc and objdump and just makes the output assembly code harder to read.
This patch changes things to always print both operands making us consistent with gcc and objdump. The parser should still be able to handle the single register forms just as it did before. This also matches the GNU assembler behavior.
llvm-svn: 353061
|
|
|
|
|
|
|
|
| |
printing it as %st(0) when its encoded in the instruction.
This is a step back from the change I made in r352985. This appears to be more consistent with gcc and objdump behavior.
llvm-svn: 353015
|
|
|
|
|
|
|
|
|
|
| |
as the clobber name to make MS inline asm work correctly"
Looking into gcc and objdump behavior more this was overly aggressive. If the register is encoded in the instruction we should print %st(0), if its implicit we should print %st.
I'll be making a more directed change in a future patch.
llvm-svn: 353013
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
name to make MS inline asm work correctly
Summary:
When calculating clobbers for MS style inline assembly we fail if the asm clobbers stack top because we print st(0) and try to pass it through the gcc register name check. This was found with when I attempted to make a emms/femms clobber all ST registers. If you use emms/femms in MS inline asm we would try to use st(0) as the clobber name but clang would think that wasn't a valid clobber name.
This also matches what objdump disassembly prints. It's also what is printed by gcc -S.
Reviewers: RKSimon, rnk, efriedma, spatel, andreadb, lebedev.ri
Reviewed By: rnk
Subscribers: eraman, gbedwell, lebedev.ri, llvm-commits
Differential Revision: https://reviews.llvm.org/D57621
llvm-svn: 352985
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
conversions.
Account for bypass delays when computing the latency of scalar int-to-float
conversions.
On Jaguar we need to account for an extra 6cy latency (see AMD fam16h SOG).
This patch also fixes the number of micropcodes for the register-memory variants
of scalar int-to-float conversions.
Differential Revision: https://reviews.llvm.org/D57148
llvm-svn: 352518
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
delays from the integer to the floating point unit.
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
|
|
|
|
|
|
| |
and rdtsc/rdtscp tests
llvm-svn: 351835
|
|
|
|
| |
llvm-svn: 351831
|
|
|
|
|
|
| |
These technically should be under a MONITOR cpuid bit, but we tag them as SSE3 so I've done that here as well.
llvm-svn: 351829
|
|
|
|
| |
llvm-svn: 351827
|
|
|
|
| |
llvm-svn: 351822
|
|
|
|
|
|
|
|
| |
Similar to horizontal ops on D56777, the sse2 (but not mmx) bit shift ops has local forwarding disabled, adding +1cy to the use latency for the result.
Differential Revision: https://reviews.llvm.org/D57026
llvm-svn: 351817
|
|
|
|
|
|
|
|
| |
Similar to horizontal ops on D56777, the vpermilpd/vpermilps variable mask ops has local forwarding disabled, adding +1cy to the use latency for the result.
Differential Revision: https://reviews.llvm.org/D57022
llvm-svn: 351815
|
|
|
|
|
|
|
|
| |
D56777 added +1cy local forwarding penalty for horizontal operations, but this penalty only affects sse2/xmm variants, the mmx variants don't suffer the penalty.
Confirmed with @andreadb
llvm-svn: 351755
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
r327630 introduced new write definitions for float/vector loads.
Before that revision, WriteLoad was used by both integer/float (scalar/vector)
load. So, WriteLoad had to conservatively declare a latency to 5cy. That is
because the load-to-use latency for float/vector load is 5cy.
Now that we have dedicated writes for float/vector loads, there is no reason why
we should keep the latency of WriteLoad to 5cy. At the moment, WriteLoad is only
used by scalar integer loads only; we can assume an optimstic 3cy latency for
them.
This patch changes that latency from 5cy to 3cy, and regenerates the affected
scheduling/mca tests.
Differential Revision: https://reviews.llvm.org/D56922
llvm-svn: 351742
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
On Jaguar, horizontal adds/subs have local forwarding disable.
That means, we pay a compulsory extra cycle of write-back stage, and the value
is not available until the end of that stage.
This patch changes the latency of horizontal operations by adding an extra
cycle. With this patch, latency numbers now match what is reported by perf.
I plan to send another patch to also 'fix' the latency of shuffle operations (on
Jaguar, local forwarding is disabled for vector shuffles too).
Differential Revision: https://reviews.llvm.org/D56777
llvm-svn: 351366
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RetireControlUnitStatistics now reports extra information about the ROB and the
avg/maximum number of entries consumed over the entire simulation.
Example:
Retire Control Unit - number of cycles where we saw N instructions retired:
[# retired], [# cycles]
0, 109 (17.9%)
1, 102 (16.7%)
2, 399 (65.4%)
Total ROB Entries: 64
Max Used ROB Entries: 35 ( 54.7% )
Average Used ROB Entries per cy: 32 ( 50.0% )
Documentation in llvm/docs/CommandGuide/llvmn-mca.rst has been updated to
reflect this change.
llvm-svn: 347493
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes an invalid memory read introduced by r346487.
Before this patch, partial register write had to query the latency of the
dependent full register write by calling a method on the full write descriptor.
However, if the full write is from an already retired instruction, chances are
that the EntryStage already reclaimed its memory.
In some parial register write tests, valgrind was reporting an invalid
memory read.
This change fixes the invalid memory access problem. Writes are now responsible
for tracking dependent partial register writes, and notify them in the event of
instruction issued.
That means, partial register writes no longer need to query their associated
full write to check when they are ready to execute.
Added test X86/BtVer2/partial-reg-update-7.s
llvm-svn: 347459
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
RegisterFileStatistics.
This patch teaches view RegisterFileStatistics how to report events for
optimizable register moves.
For each processor register file, view RegisterFileStatistics reports the
following extra information:
- Number of optimizable register moves
- Number of register moves eliminated
- Number of zero moves (i.e. register moves that propagate a zero)
- Max Number of moves eliminated per cycle.
Differential Revision: https://reviews.llvm.org/D53976
llvm-svn: 345865
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
via tablegen.
This patch adds the ability to identify instructions that are "move elimination
candidates". It also allows scheduling models to describe processor register
files that allow move elimination.
A move elimination candidate is an instruction that can be eliminated at
register renaming stage.
Each subtarget can specify which instructions are move elimination candidates
with the help of tablegen class "IsOptimizableRegisterMove" (see
llvm/Target/TargetInstrPredicate.td).
For example, on X86, BtVer2 allows both GPR and MMX/SSE moves to be eliminated.
The definition of 'IsOptimizableRegisterMove' for BtVer2 looks like this:
```
def : IsOptimizableRegisterMove<[
InstructionEquivalenceClass<[
// GPR variants.
MOV32rr, MOV64rr,
// MMX variants.
MMX_MOVQ64rr,
// SSE variants.
MOVAPSrr, MOVUPSrr,
MOVAPDrr, MOVUPDrr,
MOVDQArr, MOVDQUrr,
// AVX variants.
VMOVAPSrr, VMOVUPSrr,
VMOVAPDrr, VMOVUPDrr,
VMOVDQArr, VMOVDQUrr
], CheckNot<CheckSameRegOperand<0, 1>> >
]>;
```
Definitions of IsOptimizableRegisterMove from processor models of a same
Target are processed by the SubtargetEmitter to auto-generate a target-specific
override for each of the following predicate methods:
```
bool TargetSubtargetInfo::isOptimizableRegisterMove(const MachineInstr *MI)
const;
bool MCInstrAnalysis::isOptimizableRegisterMove(const MCInst &MI, unsigned
CPUID) const;
```
By default, those methods return false (i.e. conservatively assume that there
are no move elimination candidates).
Tablegen class RegisterFile has been extended with the following information:
- The set of register classes that allow move elimination.
- Maxium number of moves that can be eliminated every cycle.
- Whether move elimination is restricted to moves from registers that are
known to be zero.
This patch is structured in three part:
A first part (which is mostly boilerplate) adds the new
'isOptimizableRegisterMove' target hooks, and extends existing register file
descriptors in MC by introducing new fields to describe properties related to
move elimination.
A second part, uses the new tablegen constructs to describe move elimination in
the BtVer2 scheduling model.
A third part, teaches llm-mca how to query the new 'isOptimizableRegisterMove'
hook to mark instructions that are candidates for move elimination. It also
teaches class RegisterFile how to describe constraints on move elimination at
PRF granularity.
llvm-mca tests for btver2 show differences before/after this patch.
Differential Revision: https://reviews.llvm.org/D53134
llvm-svn: 344334
|
|
|
|
| |
llvm-svn: 344253
|
|
|
|
|
|
|
| |
These should test all the optimizable moves on Jaguar.
A follow-up patch will teach how to recognize these optimizable register moves.
llvm-svn: 344144
|