diff options
author | Alon Zakai <azakai@google.com> | 2023-10-03 16:39:12 -0700 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-10-03 16:39:12 -0700 |
commit | b2e096d79c36daa2cbfb7dc3db31af76e9f45cc8 (patch) | |
tree | 52f962128c9fe6c1b8cc8b847e5af40f33117c40 /test/lit/passes | |
parent | 24779b2a3fe5e5c7cc6b1da3661d346cd9c129ae (diff) | |
download | binaryen-b2e096d79c36daa2cbfb7dc3db31af76e9f45cc8.tar.gz binaryen-b2e096d79c36daa2cbfb7dc3db31af76e9f45cc8.tar.bz2 binaryen-b2e096d79c36daa2cbfb7dc3db31af76e9f45cc8.zip |
RemoveUnusedBrs: Allow less unconditional work and in particular division (#5989)
Fixes #5983: The testcase from there is used here in a new testcase
remove-unused-brs_levels in which we check if we are willing to unconditionally
do a division operation. Turning an if with an arm that does a division into a
select, which always does the division, is almost 5x slower, so we should probably
be extremely careful about doing that.
I took some measurements and have some suggestions for changes in this PR:
* Raise the cost of div/rem to what I measure on my machine, which is 5x slower
than an add, or worse.
* For some reason we added the if arms rather than take the max of them, so
fix that. This does not help the issue, but was confusing.
* Adjust TooCostlyToRunUnconditionally in the pass from 9 to 8 (this helps
balance the last point).
* Use half that value when not optimizing for size. That is, we allow only 4 extra
unconditional work normally, and 8 in -Os, and when -Oz then we allow any
extra amount.
Aside from the new testcases, some existing ones changed. They all appear to
change in a reasonable way, to me.
We should perhaps go even further than this, and not even run a division
unconditionally in -Os, but I wasn't sure it makes sense to go that far as
other benchmarks may be affected. For now, this makes the benchmark in
#5983 run at full speed in -O3 or -Os, and it remains slow in -Oz. The
modified version of the benchmark that only divides in the if (no other
operations) is still fast in -O3, but it become slow in -Os as we do turn that
if into a select (but again, I didn't want to go that far as to overfit on that one
benchmark).
Diffstat (limited to 'test/lit/passes')
-rw-r--r-- | test/lit/passes/remove-unused-brs-gc.wast | 22 | ||||
-rw-r--r-- | test/lit/passes/remove-unused-brs.wast | 10 | ||||
-rw-r--r-- | test/lit/passes/remove-unused-brs_levels.wast | 129 |
3 files changed, 142 insertions, 19 deletions
diff --git a/test/lit/passes/remove-unused-brs-gc.wast b/test/lit/passes/remove-unused-brs-gc.wast index 53100cc91..7a620193e 100644 --- a/test/lit/passes/remove-unused-brs-gc.wast +++ b/test/lit/passes/remove-unused-brs-gc.wast @@ -655,22 +655,20 @@ ;; CHECK-NEXT: ) ;; CHECK-NEXT: ) ;; CHECK-NEXT: (drop - ;; CHECK-NEXT: (select (result (ref null $struct)) - ;; CHECK-NEXT: (block (result (ref null $struct)) - ;; CHECK-NEXT: (block $something (result (ref null $struct)) - ;; CHECK-NEXT: (drop - ;; CHECK-NEXT: (block (result nullref) - ;; CHECK-NEXT: (br_on_non_null $something - ;; CHECK-NEXT: (local.get $struct) - ;; CHECK-NEXT: ) - ;; CHECK-NEXT: (ref.null none) + ;; CHECK-NEXT: (if (result (ref null $struct)) + ;; CHECK-NEXT: (local.get $x) + ;; CHECK-NEXT: (block $something (result (ref null $struct)) + ;; CHECK-NEXT: (drop + ;; CHECK-NEXT: (block (result nullref) + ;; CHECK-NEXT: (br_on_non_null $something + ;; CHECK-NEXT: (local.get $struct) ;; CHECK-NEXT: ) + ;; CHECK-NEXT: (ref.null none) ;; CHECK-NEXT: ) - ;; CHECK-NEXT: (ref.null none) ;; CHECK-NEXT: ) + ;; CHECK-NEXT: (ref.null none) ;; CHECK-NEXT: ) ;; CHECK-NEXT: (ref.null none) - ;; CHECK-NEXT: (local.get $x) ;; CHECK-NEXT: ) ;; CHECK-NEXT: ) ;; CHECK-NEXT: (drop @@ -716,6 +714,8 @@ ) ) ) + ;; We do not selectify here because the amount of work in the if is + ;; significant (there is a cast and a branch). (drop (if (result anyref) (local.get $x) diff --git a/test/lit/passes/remove-unused-brs.wast b/test/lit/passes/remove-unused-brs.wast index 82963f792..93cf4cbd2 100644 --- a/test/lit/passes/remove-unused-brs.wast +++ b/test/lit/passes/remove-unused-brs.wast @@ -31,10 +31,7 @@ ;; CHECK-NEXT: (i32.const 1) ;; CHECK-NEXT: (i32.lt_u ;; CHECK-NEXT: (i32.sub - ;; CHECK-NEXT: (i32.or - ;; CHECK-NEXT: (local.get $0) - ;; CHECK-NEXT: (i32.const 32) - ;; CHECK-NEXT: ) + ;; CHECK-NEXT: (local.get $0) ;; CHECK-NEXT: (i32.const 97) ;; CHECK-NEXT: ) ;; CHECK-NEXT: (i32.const 6) @@ -60,10 +57,7 @@ (i32.const 1) (i32.lt_u (i32.sub - (i32.or - (local.get $0) - (i32.const 32) - ) + (local.get $0) (i32.const 97) ) (i32.const 6) diff --git a/test/lit/passes/remove-unused-brs_levels.wast b/test/lit/passes/remove-unused-brs_levels.wast new file mode 100644 index 000000000..f4737627e --- /dev/null +++ b/test/lit/passes/remove-unused-brs_levels.wast @@ -0,0 +1,129 @@ +;; NOTE: Assertions have been generated by update_lit_checks.py and should not be edited. +;; RUN: wasm-opt %s --remove-unused-brs -all -S --shrink-level=0 -o - | filecheck %s --check-prefix=SHRINK_0 +;; RUN: wasm-opt %s --remove-unused-brs -all -S --shrink-level=1 -o - | filecheck %s --check-prefix=SHRINK_1 +;; RUN: wasm-opt %s --remove-unused-brs -all -S --shrink-level=2 -o - | filecheck %s --check-prefix=SHRINK_2 + + +(module + ;; SHRINK_0: (func $selectify-division (type $0) (param $x i32) (result i32) + ;; SHRINK_0-NEXT: (if (result i32) + ;; SHRINK_0-NEXT: (i32.eq + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: (i32.const 53498923) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: (i32.div_s + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: (i32.const 13) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_1: (func $selectify-division (type $0) (param $x i32) (result i32) + ;; SHRINK_1-NEXT: (select + ;; SHRINK_1-NEXT: (i32.div_s + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: (i32.const 13) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: (i32.eq + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: (i32.const 53498923) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_2: (func $selectify-division (type $0) (param $x i32) (result i32) + ;; SHRINK_2-NEXT: (select + ;; SHRINK_2-NEXT: (i32.div_s + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.const 13) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.eq + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.const 53498923) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: ) + (func $selectify-division (param $x i32) (result i32) + ;; See #5983: this if, if turned into a select, becomes almost 5x slower. + ;; We only want to selectify here when the shrink level is 1 or 2. + (if (result i32) + (i32.eq + (local.get $x) + (i32.const 53498923) + ) + (i32.div_s + (local.get $x) + (i32.const 13) + ) + (local.get $x) + ) + ) + + ;; SHRINK_0: (func $selectify-division2 (type $0) (param $x i32) (result i32) + ;; SHRINK_0-NEXT: (if (result i32) + ;; SHRINK_0-NEXT: (i32.eq + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: (i32.const 53498923) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: (i32.div_s + ;; SHRINK_0-NEXT: (i32.div_s + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: (i32.const 13) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: (i32.const 13) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: (local.get $x) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_0-NEXT: ) + ;; SHRINK_1: (func $selectify-division2 (type $0) (param $x i32) (result i32) + ;; SHRINK_1-NEXT: (if (result i32) + ;; SHRINK_1-NEXT: (i32.eq + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: (i32.const 53498923) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: (i32.div_s + ;; SHRINK_1-NEXT: (i32.div_s + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: (i32.const 13) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: (i32.const 13) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: (local.get $x) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_1-NEXT: ) + ;; SHRINK_2: (func $selectify-division2 (type $0) (param $x i32) (result i32) + ;; SHRINK_2-NEXT: (select + ;; SHRINK_2-NEXT: (i32.div_s + ;; SHRINK_2-NEXT: (i32.div_s + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.const 13) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: (i32.const 13) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.eq + ;; SHRINK_2-NEXT: (local.get $x) + ;; SHRINK_2-NEXT: (i32.const 53498923) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: ) + ;; SHRINK_2-NEXT: ) + (func $selectify-division2 (param $x i32) (result i32) + ;; As above, but now only with a shrink level of 2 should we selectify, as + ;; there are two divisions. + (if (result i32) + (i32.eq + (local.get $x) + (i32.const 53498923) + ) + (i32.div_s + (i32.div_s + (local.get $x) + (i32.const 13) + ) + (i32.const 13) + ) + (local.get $x) + ) + ) +) |