Skip to content

[OptimizeInstructions] Combine extend to i64 and 32-bit load operations #4307

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 17 commits into from
Nov 12, 2021
Merged

[OptimizeInstructions] Combine extend to i64 and 32-bit load operations #4307

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
a48712b
init
MaxGraey Nov 5, 2021
7538df6
exclude two special cases
MaxGraey Nov 5, 2021
1b44333
refactor comments
MaxGraey Nov 5, 2021
e01e9ea
Merge branch 'main' into combine-load32-store32-zero-ext-64
MaxGraey Nov 5, 2021
d41136d
Merge branch 'main' into combine-load32-store32-zero-ext-64
MaxGraey Nov 10, 2021
3e84878
lift isAtomic check up
MaxGraey Nov 10, 2021
6ec5c89
simplify more
MaxGraey Nov 10, 2021
e807d3b
more
MaxGraey Nov 10, 2021
99a4385
fix
MaxGraey Nov 10, 2021
6b81ffd
allow this for atomics as well
MaxGraey Nov 10, 2021
0892bb3
add comment about special cases
MaxGraey Nov 10, 2021
4efb6bc
refactor comment
MaxGraey Nov 10, 2021
e702d9f
simplify
MaxGraey Nov 10, 2021
e9df929
fix
MaxGraey Nov 10, 2021
6259658
fix special case for sext(i64.atomic.load(x)). Add tests for atomics
MaxGraey Nov 12, 2021
0245cc2
refactor
MaxGraey Nov 12, 2021
b07c84b
move comment
MaxGraey Nov 12, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 40 additions & 0 deletions src/passes/OptimizeInstructions.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -907,6 +907,46 @@ struct OptimizeInstructions
}
}

if (curr->op == ExtendUInt32 || curr->op == ExtendSInt32) {
if (auto* load = curr->value->dynCast<Load>()) {
// i64.extend_i32_s(i32.load(_8|_16)(_u|_s)(x)) =>
// i64.load(_8|_16|_32)(_u|_s)(x)
//
// i64.extend_i32_u(i32.load(_8|_16)(_u|_s)(x)) =>
// i64.load(_8|_16|_32)(_u|_s)(x)
//
// but we can't do this in following cases:
//
// i64.extend_i32_u(i32.load8_s(x))
// i64.extend_i32_u(i32.load16_s(x))
//
// this mixed sign/zero extensions can't represent in single
// signed or unsigned 64-bit load operation. For example if `load8_s(x)`
// return i8(-1) (0xFF) than sign extended result will be
// i32(-1) (0xFFFFFFFF) and with zero extension to i64 we got
// finally 0x00000000FFFFFFFF. However with `i64.load8_s` in this
// situation we got `i64(-1)` (all ones) and with `i64.load8_u` it
// will be 0x00000000000000FF.
//
// Another limitation is atomics which only have unsigned loads.
// So we also avoid this only case:
//
// i64.extend_i32_s(i32.atomic.load(x))

// Special case for i32.load. In this case signedness depends on
// extend operation.
bool willBeSigned = curr->op == ExtendSInt32 && load->bytes == 4;
if (!(curr->op == ExtendUInt32 && load->bytes <= 2 && load->signed_) &&
!(willBeSigned && load->isAtomic)) {
if (willBeSigned) {
load->signed_ = true;
}
load->type = Type::i64;
return replaceCurrent(load);
}
}
}

if (Abstract::hasAnyReinterpret(curr->op)) {
// i32.reinterpret_f32(f32.reinterpret_i32(x)) => x
// i64.reinterpret_f64(f64.reinterpret_i64(x)) => x
Expand Down
44 changes: 44 additions & 0 deletions test/lit/passes/optimize-instructions-atomics.wast
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,4 +51,48 @@
(drop (f64.reinterpret_i64 (i64.atomic.load (local.get $x)))) ;; skip
(i32.atomic.store (i32.const 8) (i32.reinterpret_f32 (local.get $y))) ;; skip
)

;; CHECK: (func $combine_atomic_load_and_extends (param $x i32)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.atomic.load8_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.atomic.load16_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.atomic.load32_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.atomic.load8_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.atomic.load16_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.extend_i32_s
;; CHECK-NEXT: (i32.atomic.load
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
(func $combine_atomic_load_and_extends (param $x i32)
(drop (i64.extend_i32_u (i32.atomic.load8_u (local.get $x))))
(drop (i64.extend_i32_u (i32.atomic.load16_u (local.get $x))))
(drop (i64.extend_i32_u (i32.atomic.load (local.get $x))))
(drop (i64.extend_i32_s (i32.atomic.load8_u (local.get $x))))
(drop (i64.extend_i32_s (i32.atomic.load16_u (local.get $x))))
;; skips
(drop (i64.extend_i32_s (i32.atomic.load (local.get $x))))
)
)
83 changes: 83 additions & 0 deletions test/lit/passes/optimize-instructions.wast
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13404,4 +13404,87 @@
(drop (f32.reinterpret_i32 (i32.reinterpret_f32 (local.get $z))))
(drop (f64.reinterpret_i64 (i64.reinterpret_f64 (local.get $w))))
)

;; u64(i32.load(_8|_16)(_u|_s)(x)) => i64.load(_8|_16|_32)(_u|_s)(x)
;; except:
;; i64.extend_i32_u(i32.load8_s(x)) and
;; i64.extend_i32_u(i32.load16_s(x))

;; CHECK: (func $combine_load_and_extend_u (param $x i32)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load8_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load16_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load32_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.extend_i32_u
;; CHECK-NEXT: (i32.load8_s
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.extend_i32_u
;; CHECK-NEXT: (i32.load16_s
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
(func $combine_load_and_extend_u (param $x i32)
(drop (i64.extend_i32_u (i32.load8_u (local.get $x))))
(drop (i64.extend_i32_u (i32.load16_u (local.get $x))))
(drop (i64.extend_i32_u (i32.load (local.get $x))))

;; skips
(drop (i64.extend_i32_u (i32.load8_s (local.get $x))))
(drop (i64.extend_i32_u (i32.load16_s (local.get $x))))
)

;; i64(i32.load(_8|_16)(_u|_s)(x)) => i64.load(_8|_16|_32)(_u|_s)(x)

;; CHECK: (func $combine_load_and_extend_s (param $x i32)
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load8_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load16_u
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load8_s
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load16_s
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: (drop
;; CHECK-NEXT: (i64.load32_s
;; CHECK-NEXT: (local.get $x)
;; CHECK-NEXT: )
;; CHECK-NEXT: )
;; CHECK-NEXT: )
(func $combine_load_and_extend_s (param $x i32)
(drop (i64.extend_i32_s (i32.load8_u (local.get $x))))
(drop (i64.extend_i32_s (i32.load16_u (local.get $x))))
(drop (i64.extend_i32_s (i32.load8_s (local.get $x))))
(drop (i64.extend_i32_s (i32.load16_s (local.get $x))))
(drop (i64.extend_i32_s (i32.load (local.get $x))))
)
)