* Avoid calling memmove(dst, nullptr, 0) as this is flagged by UBSAN.
* Avoid hitting a bug[1] in the linker: LLD's identical code folding
(ICF) happens to replace RecordCoverageInstr::DebugName() with
DispatchTable::LargestSmallOffset() because they happen to contain
the same machine code, ICF fails to accomodate that DebugName also
contains a relocation to constant string. To avoid this we simply
eliminate LargestSmallOffset and replace it with a constant. Same for
OriginElement.
TEST=manually tested previously failing tests
[1]: reported https://github.com/llvm/llvm-project/issues/57693
Change-Id: I38637df6475c7670081b7af0a2de75ca37f6f07c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/258801
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Commit-Queue: Slava Egorov <vegorov@google.com>
Implements a backend targeting RV32GC and RV64GC, based on Linux standardizing around GC. The assembler is written to make it easy to disable usage of C, but because the sizes of some instruction sequences are compile-time constants, an additional build configuration would need to be defined to make use of it.
The assembler and disassembler cover every RV32/64GC instruction. The simulator covers all instructions except accessing CSRs and the floating point state accessible through such, include accrued exceptions and dynamic rounding mode.
Quirks:
- RISC-V is a compare-and-branch architecture, but some existing "architecture-independent" parts of the Dart compiler assume a condition code architecture. To avoid rewriting these parts, we use a peephole in the assembler to map to compare-and-branch. See Assembler::BranchIf. Luckily nothing depended on taking multiple branches on the same condition code set.
- There are no hardware overflow checks, so we must use Hacker's Delight style software checks. Often these are very cheap: if the sign of one operand is known, a single branch is needed.
- The ranges of RISC-V branches and jumps are such that we use 3 levels of generation for forward branches, instead of the 2 levels of near and far branches used on ARM[64]. Nearly all code is handled by the first two levels with 20-bits of range, with enormous regex matchers triggering the third level that uses aupic+jalr to get 32-bits of range.
- For PC-relative calls in AOT, we always generate auipc+jalr pairs with 32-bits of range, so we never generate trampolines.
- Only a subset of registers are available in some compressed instructions, so we assign the most popular uses to these registers. In particular, THR, TMP[2], CODE and PP. This has the effect of assigning CODE and PP to volatile registers in the C calling convention, whereas they are assigned preserved registers on the other architectures. As on ARM64, PP is untagged; this is so short indices can be accessed with a compressed instruction.
- There are no push or pop instructions, so combining pushes and pops is preferred so we can update SP once.
- The C calling convention has a strongly aligned stack, but unlike on ARM64 we don't need to use an alternate stack pointer. The author ensured language was added to the RISC-V psABI making the OS responsible for realigning the stack pointer for signal handlers, allowing Dart to leave the stack pointer misaligned from the C calling convention's point of view until a foreign call.
- We don't bother with the link register tracking done on ARM[64]. Instead we make use of an alternate link register to avoid inline spilling in the write barrier.
Unimplemented:
- non-trivial FFI cases
- Compressed pointers - No intention to implement.
- Unboxed SIMD - We might make use of the V extension registers when the V extension is ratified.
- BigInt intrinsics
TEST=existing tests for IL level, new tests for assembler/disassembler/simulator
Bug: https://github.com/dart-lang/sdk/issues/38587
Bug: https://github.com/dart-lang/sdk/issues/48164
Change-Id: I991d1df4be5bf55efec5371b767b332d37dfa3e0
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/217289
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Daco Harkes <dacoharkes@google.com>
Reviewed-by: Slava Egorov <vegorov@google.com>
Commit-Queue: Ryan Macnak <rmacnak@google.com>
These changes were originally submitted separately on different days,
and a major performance regression was seen after the first change
when creating snapshots that led to both being reverted. However,
that performance regression should be addressed by the followup.
First change:
"[vm] Treat the dispatch table as a root in the snapshot.
Additional changes:
* Only serialize a dispatch table in precompiled snapshots.
* Add information in v8 snapshot profiles for the dispatch table.
* Fix a typo in a field name.
* Print the number of Instructions objects (or payloads, for
precompiled bare instructions mode) in the fake cluster for
the data section.
* Fix v8 snapshots profiles so objects in memory mapped segments
and only those are prefixed with "(RO) ".
* Add names for Instructions objects in v8 snapshot profiles
when we can use the assembly namer.
* Add command line flag for old #define'd false flag."
Second change:
"[vm/aot] Keep GC-visible references to dispatch table Code entries.
This change splits dispatch table handling into four distinct
parts:
* The dispatch table generator does not make a dispatch table
directly, but rather creates an Array that contains the Code
objects for dispatch table entries.
* The precompiler takes this Array and puts it in the object
store, which makes it a new GC root.
* The serializer takes this information and serializes the
dispatch table information in the same form as before.
* The deserializer creates a DispatchTable object and populates
it using the serialized information.
The change in the precompiler ensures that the Code objects
used in the dispatch table have GC-visible references. Thus,
even if all other references to them from the other GC roots
were removed, they would be accessible in the serializer in
the case of a GC pass between the precompiler and serializer.
This change also means that the serializer can retrieve and
trace the Code objects directly rather than first looking up
the Code objects by their entry point."
Bug: https://github.com/dart-lang/sdk/issues/41022
Change-Id: I52c83b0536fc588da0bef9aed1f0c72e8ee4663f
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm_x64-try,vm-kernel-precomp-android-release-arm64-try,vm-kernel-precomp-android-release-arm_x64-try,vm-kernel-precomp-mac-release-simarm64-try,vm-kernel-precomp-win-release-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/139285
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Alexander Aprelev <aam@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This change splits dispatch table handling into four distinct
parts:
* The dispatch table generator does not make a dispatch table
directly, but rather creates an Array that contains the Code
objects for dispatch table entries.
* The precompiler takes this Array and puts it in the object
store, which makes it a new GC root.
* The serializer takes this information and serializes the
dispatch table information in the same form as before.
* The deserializer creates a DispatchTable object and populates
it using the serialized information.
The change in the precompiler ensures that the Code objects
used in the dispatch table have GC-visible references. Thus,
even if all other references to them from the other GC roots
were removed, they would be accessible in the serializer in
the case of a GC pass between the precompiler and serializer.
This change also means that the serializer can retrieve and
trace the Code objects directly rather than first looking up
the Code objects by their entry point.
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-linux-release-x64-try,vm-kernel-precomp-linux-release-simarm-try,vm-kernel-precomp-linux-release-simarm64-try,vm-kernel-precomp-linux-release-simarm_x64-try,vm-kernel-precomp-android-release-arm64-try,vm-kernel-precomp-android-release-arm_x64-try,vm-kernel-precomp-mac-release-simarm64-try,vm-kernel-precomp-win-release-x64-try
Change-Id: Ic9ba29dfd64945a6e9a51f9de73f74d9202b58b7
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/138288
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Aske Simon Christensen <askesc@google.com>
NullCheck pc descriptors are added to all dispatch table calls where the
receiver may be null (and the selector is not one implemented by null).
All null entries in the table go to the NullError runtime entry, which
reads the NullCheck pc descriptor to get the name of the called member
for the error message.
Change-Id: I9d2847d0ccdfdb735b06e879916920ec299f39bc
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/134294
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This is a Dart-tailored implementation of the "row-displacement
dispatch table" technique for closed-world instance calls:
All interface targets in the program are grouped into selectors such
that all targets that could potentially be called from the same call
site have the same selector (currently just grouped by name).
Each selector is assigned a selector offset such that offset + classid
is unique for all selector/classid combinations where the class
implements the selector.
At every instance call site that has an interface target (i.e. where
the static type of the receiver is not dynamic), the selector offset +
receiver classid is computed and used as index into a global table of
entry points.
If the receiver can be null (as determined by the front-end TFA and the
VM type propagation), a null check is inserted before the call.
An arguments descriptor is provided (only) for selectors that need it
(those which have type parameters or optional/named parameters).
The dispatch table calls don't need the monomorphic entry code, so for
functions that are only called via dispatch table calls (i.e. never
called dynamically), the monomorphic entry code is left out.
Some future improvements to the table dispatch implementation are
mentioned in https://github.com/dart-lang/sdk/issues/40188
The table dispatch flag is disabled by default in this commit. A
separate commit enables the flag.
Change-Id: Ic2911742b4a2c9a8d3bc7df60605454cbe4c0714
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/126648
Commit-Queue: Aske Simon Christensen <askesc@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Vyacheslav Egorov