Stream: git-wasmtime

Topic: wasmtime / PR #2565 Detailed debug-info (DWARF) support i...

Wasmtime GitHub notifications bot (Jan 09 2021 at 21:05):

cfallin edited PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 09 2021 at 21:44):

cfallin updated PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 10 2021 at 03:21):

cfallin updated PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 21 2021 at 22:06):

cfallin requested fitzgen, bnjbvr and yurydelendik for a review on PR #2565.

Wasmtime GitHub notifications bot (Jan 21 2021 at 22:06):

cfallin requested fitzgen, bnjbvr and yurydelendik for a review on PR #2565.

Wasmtime GitHub notifications bot (Jan 22 2021 at 00:01):

cfallin updated PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 22 2021 at 14:15):

yurydelendik submitted PR Review.

Wasmtime GitHub notifications bot (Jan 22 2021 at 14:15):

yurydelendik created PR Review Comment:

ExpressionWriter has write_op_breg

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen submitted PR Review.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

All of these methods could use a little, single-sentence doc comment describing what they're doing.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen submitted PR Review.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

Is the worklist processing order important here (eg avoiding big O blow up)? If not, I suspect that Vec and pop would be faster than VecDeque and pop_front.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

What is remove here? Can you add comments for this struct and its members?

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

Ah is this when a collection becomes empty and can be removed from its parent collection? Maybe is_empty is a better name, or maybe just having a comment that explains would be enough.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

This will always search for stack even if reg.is_some(). This can be remedied by using .or_else(inline_the_def_of_stack_here).

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

This could be a little tighter as

if list.last().map_or(true, |last| last.end <= offset || last.loc != loc) {
    ...
}

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

What is this? I have no idea what iix means here.

Wasmtime GitHub notifications bot (Jan 22 2021 at 23:00):

fitzgen created PR Review Comment:

remove_keys.clear() to reuse the previous allocation

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:02):

cfallin updated PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin created PR Review Comment:

The map_or(true, ...) form is a little hard to read for me, so I did list.last().map(...).unwrap_or(true) to be a little more explicit, and added a comment too. Happy to make things more concise, just want to avoid Haskell-like terseness that takes too much brain (for me, anyway) to decode :-)

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin created PR Review Comment:

Nice improvement, thanks!

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:03):

cfallin created PR Review Comment:

Done; no, it shouldn't be important.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin created PR Review Comment:

Changed to index; it's a regalloc.rs-ism but I should really just write it out.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin created PR Review Comment:

Done, thanks!

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin created PR Review Comment:

Added comment and renamed; agreed, that's more clear.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:04):

cfallin created PR Review Comment:

Done; sorry about the terseness!

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:45):

cfallin updated PR #2565 from debug-value-labels to main:

This PR propagates "value labels" all the way from CLIF to DWARF
metadata on the emitted machine code. The key idea is as follows:

• Translate value-label metadata on the input into "value_label"
  pseudo-instructions when lowering into VCode. These
  pseudo-instructions take a register as input, denote a value label,
  and semantically are like a "move into value label" -- i.e., they
  update the current value (as seen by debugging tools) of the given
  local. These pseudo-instructions emit no machine code.

• Perform a dataflow analysis at the machine-code level, tracking
  value-labels that propagate into registers and into [SP+constant]
  stack storage. This is a forward dataflow fixpoint analysis where each
  storage location can contain a set of value labels, and each value
  label can reside in a set of storage locations. (Meet function is
  pairwise intersection by storage location.)

  This analysis traces value labels symbolically through loads and
  stores and reg-to-reg moves, so it will naturally handle spills and
  reloads without knowing anything special about them.

• When this analysis converges, we have, at each machine-code offset, a
  mapping from value labels to some number of storage locations; for
  each offset for each label, we choose the best location (prefer
  registers). Note that we can choose any location, as the symbolic
  dataflow analysis is sound and guarantees that the value at the
  value_label instruction propagates to all of the named locations.

• Then we can convert this mapping into a format that the DWARF
  generation code (wasmtime's debug crate) can use.

This PR also adds the new-backend variant to the gdb tests on CI.

Wasmtime GitHub notifications bot (Jan 23 2021 at 00:56):

fitzgen submitted PR Review.

Wasmtime GitHub notifications bot (Jan 23 2021 at 01:22):

cfallin merged PR #2565.

Last updated: Dec 23 2024 at 13:07 UTC