cranelift-isle: Add "partial" flag for constructors (#5392)

* cranelift-isle: Add "partial" flag for constructors

Instead of tying fallibility of constructors to whether they're either
internal or pure, this commit assumes all constructors are infallible
unless tagged otherwise with a "partial" flag.

Internal constructors without the "partial" flag are not allowed to use
constructors which have the "partial" flag on the right-hand side of any
rules, because they have no way to report last-minute match failures.

Multi-constructors should never be "partial"; they report match failures
with an empty iterator instead. In turn this means you can't use partial
constructors on the right-hand side of internal multi-constructor rules.
However, you can use the same constructors on the left-hand side with
`if` or `if-let` instead.

In many cases, ISLE can already trivially prove that an internal
constructor always returns `Some`. With this commit, those cases are
largely unchanged, except for removing all the `Option`s and `Some`s
from the generated code for those terms.

However, for internal non-partial constructors where ISLE could not
prove that, it now emits an `unreachable!` panic as the last-resort,
instead of returning `None` like it used to do. Among the existing
backends, here's how many constructors have these panic cases:

- x64: 14% (53/374)
- aarch64: 15% (41/277)
- riscv64: 23% (26/114)
- s390x: 47% (268/567)

It's often possible to rewrite rules so that ISLE can tell the panic can
never be hit. Just ensure that there's a lowest-priority rule which has
no constraints on the left-hand side.

But in many of these constructors, it's difficult to statically prove
the unhandled cases are unreachable because that's only down to
knowledge about how they're called or other preconditions.

So this commit does not try to enforce that all terms have a last-resort
fallback rule.

* Check term flags while translating expressions

Instead of doing it in a separate pass afterward.

This involved threading all the term flags (pure, multi, partial)
through the recursive `translate_expr` calls, so I extracted the flags
to a new struct so they can all be passed together.

* Validate multi-term usage

Now that I've threaded the flags through `translate_expr`, it's easy to
check this case too, so let's just do it.

* Extract `ReturnKind` to use in `ExternalSig`

There are only three legal states for the combination of `multi` and
`infallible`, so replace those fields of `ExternalSig` with a
three-state enum.

* Remove `Option` wrapper from multi-extractors too

If we'd had any external multi-constructors this would correct their
signatures as well.

* Update ISLE tests

* Tag prelude constructors as pure where appropriate

I believe the only reason these weren't marked `pure` before was because
that would have implied that they're also partial. Now that those two
states are specified separately we apply this flag more places.

* Fix my changes to aarch64 `lower_bmask` and `imm` terms
This commit is contained in:
Jamey Sharp
2022-12-07 17:16:03 -08:00
committed by GitHub
parent c9527e0af6
commit 8726eeefb3
26 changed files with 433 additions and 358 deletions

View File

@@ -1537,15 +1537,15 @@
;; Detect specific integer values
(decl pure i64_nonequal (i64 i64) i64)
(decl pure partial i64_nonequal (i64 i64) i64)
(extern constructor i64_nonequal i64_nonequal)
(decl pure i64_nonzero (i64) i64)
(decl pure partial i64_nonzero (i64) i64)
(rule (i64_nonzero x)
(if (i64_nonequal x 0))
x)
(decl pure i64_not_neg1 (i64) i64)
(decl pure partial i64_not_neg1 (i64) i64)
(rule (i64_not_neg1 x)
(if (i64_nonequal x -1))
x)
@@ -1799,11 +1799,11 @@
;; Form the sum of two offset values, and check that the result is
;; a valid `MemArg::Symbol` offset (i.e. is even and fits into i32).
(decl pure memarg_symbol_offset_sum (i64 i64) i32)
(decl pure partial memarg_symbol_offset_sum (i64 i64) i32)
(extern constructor memarg_symbol_offset_sum memarg_symbol_offset_sum)
;; Likewise, but just check a single offset value.
(decl pure memarg_symbol_offset (i64) i32)
(decl pure partial memarg_symbol_offset (i64) i32)
(rule (memarg_symbol_offset x)
(memarg_symbol_offset_sum x 0))
@@ -1837,7 +1837,7 @@
;; Test whether a `load` address will be lowered to a `MemArg::Symbol`.
(decl pure load_sym (Inst) Inst)
(decl pure partial load_sym (Inst) Inst)
(rule (load_sym inst)
(if-let (load _ (symbol_value (symbol_value_data _ (reloc_distance_near) sym_offset))
(i64_from_offset load_offset))
@@ -1845,7 +1845,7 @@
(if (memarg_symbol_offset_sum sym_offset load_offset))
inst)
(decl pure uload16_sym (Inst) Inst)
(decl pure partial uload16_sym (Inst) Inst)
(rule (uload16_sym inst)
(if-let (uload16 _ (symbol_value (symbol_value_data _ (reloc_distance_near) sym_offset))
(i64_from_offset load_offset))
@@ -2752,7 +2752,7 @@
;; Similarly, because we cannot allocate temp registers, if an instruction
;; requires matching source and destination registers, this needs to be handled
;; by the user. Another helper to verify that constraint.
(decl pure same_reg (WritableReg Reg) Reg)
(decl pure partial same_reg (WritableReg Reg) Reg)
(extern constructor same_reg same_reg)
;; Push a `MInst.AluRRR` instruction to a sequence.

View File

@@ -2,12 +2,12 @@
;; The main lowering constructor term: takes a clif `Inst` and returns the
;; register(s) within which the lowered instruction's result values live.
(decl lower (Inst) InstOutput)
(decl partial lower (Inst) InstOutput)
;; A variant of the main lowering constructor term, used for branches.
;; The only difference is that it gets an extra argument holding a vector
;; of branch targets to be used.
(decl lower_branch (Inst VecMachLabel) InstOutput)
(decl partial lower_branch (Inst VecMachLabel) InstOutput)
;;;; Rules for `iconst` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

View File

@@ -526,15 +526,13 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
}
#[inline]
fn lane_order(&mut self) -> Option<LaneOrder> {
Some(lane_order_for_call_conv(
self.lower_ctx.abi().call_conv(self.lower_ctx.sigs()),
))
fn lane_order(&mut self) -> LaneOrder {
lane_order_for_call_conv(self.lower_ctx.abi().call_conv(self.lower_ctx.sigs()))
}
#[inline]
fn be_lane_idx(&mut self, ty: Type, idx: u8) -> u8 {
match self.lane_order().unwrap() {
match self.lane_order() {
LaneOrder::LittleEndian => ty.lane_count() as u8 - 1 - idx,
LaneOrder::BigEndian => idx,
}
@@ -542,7 +540,7 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
#[inline]
fn be_vec_const(&mut self, ty: Type, n: u128) -> u128 {
match self.lane_order().unwrap() {
match self.lane_order() {
LaneOrder::LittleEndian => n,
LaneOrder::BigEndian => {
let lane_count = ty.lane_count();
@@ -568,7 +566,7 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
#[inline]
fn shuffle_mask_from_u128(&mut self, idx: u128) -> (u128, u16) {
let bytes = match self.lane_order().unwrap() {
let bytes = match self.lane_order() {
LaneOrder::LittleEndian => idx.to_be_bytes().map(|x| {
if x < 16 {
15 - x
@@ -590,7 +588,7 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
let inst = self.lower_ctx.dfg().value_def(val).inst()?;
let constant = self.lower_ctx.get_constant(inst)?;
let ty = self.lower_ctx.output_ty(inst, 0);
Some(zero_extend_to_u64(constant, self.ty_bits(ty).unwrap()))
Some(zero_extend_to_u64(constant, self.ty_bits(ty)))
}
#[inline]
@@ -598,7 +596,7 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
let inst = self.lower_ctx.dfg().value_def(val).inst()?;
let constant = self.lower_ctx.get_constant(inst)?;
let ty = self.lower_ctx.output_ty(inst, 0);
Some(zero_extend_to_u64(!constant, self.ty_bits(ty).unwrap()))
Some(zero_extend_to_u64(!constant, self.ty_bits(ty)))
}
#[inline]
@@ -620,7 +618,7 @@ impl generated_code::Context for IsleContext<'_, '_, MInst, Flags, IsaFlags, 6>
let inst = self.lower_ctx.dfg().value_def(val).inst()?;
let constant = self.lower_ctx.get_constant(inst)?;
let ty = self.lower_ctx.output_ty(inst, 0);
Some(sign_extend_to_u64(constant, self.ty_bits(ty).unwrap()))
Some(sign_extend_to_u64(constant, self.ty_bits(ty)))
}
#[inline]