benchmarks: add Taylor series for pi #482
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
axic
commented
Aug 13, 2020
test/benchmarks/taylor.c
Outdated
| sum = 4.0 * sum; | ||
|
|
||
| // Display all 16 digits of double precision as a 64-bit integer | ||
| return sum * 10000000000000000ULL; |
There was a problem hiding this comment.
First I returned f64. Supporting that in wasm_engine/fizzy_engine is trivial, but changing the fizzy-bench parser seemed like a larger task so decided for this "workaround".
axic
commented
Aug 13, 2020
test/benchmarks/taylor.c
Outdated
| WASM_EXPORT unsigned long long taylor(unsigned n) | ||
| { | ||
| double sum = 1.0; | ||
| int sign = -1; |
There was a problem hiding this comment.
Could make this double too to have sign changing op, but this way there's an int to float conversion.
Alternatively make this float sign so there's a promote instruction in the loop.
axic
commented
Aug 13, 2020
| return sum; | ||
| } | ||
|
|
||
| WASM_EXPORT unsigned long long taylor_pi(unsigned n) |
There was a problem hiding this comment.
Doing this to avoid the need of f32/f64 in fizzy-bench.
axic
force-pushed
the
bench-taylor
branch
2 times, most recently
from
a468409
to
020bb75
Compare
Codecov Report
@@ Coverage Diff @@
## master #482 +/- ##
=======================================
Coverage 99.67% 99.67%
=======================================
Files 54 54
Lines 17180 17180
=======================================
Hits 17125 17125
Misses 55 55 |
axic
changed the title
chfast
previously requested changes
Aug 14, 2020
test/benchmarks/taylor_pi.inputs
Outdated
|
|
||
| 31415916535897744 | ||
|
|
||
| pi_3000000_runs |
There was a problem hiding this comment.
Please keep single execution case for a start. Something that runs in milliseconds range.
There was a problem hiding this comment.
This is on my machine:
fizzy/execute/taylor_pi/pi_1000000_runs 52600 us 51521 us 14
wabt/execute/taylor_pi/pi_1000000_runs 80834 us 79290 us 9
wasm3/execute/taylor_pi/pi_1000000_runs 12373 us 12234 us 58
fizzy/execute/taylor_pi/pi_3000000_runs 152631 us 150737 us 4
wabt/execute/taylor_pi/pi_3000000_runs 235717 us 233095 us 3
wasm3/execute/taylor_pi/pi_3000000_runs 36830 us 36516 us 19
What is your upper bound?
There was a problem hiding this comment.
Leave the first one. The second runs over 2 seconds with sanitizers on CI.
(module
(type $t0 (func))
(type $t1 (func (param i32) (result i64)))
(func $__wasm_call_ctors (type $t0))
(func $taylor_pi (export "taylor_pi") (type $t1) (param $p0 i32) (result i64)
(local $l0 i64) (local $l1 i32) (local $l2 f64) (local $l3 f64)
i64.const 40000000000000000
set_local $l0
block $B0
block $B1
get_local $p0
i32.const 2
i32.lt_u
br_if $B1
get_local $p0
i32.const -1
i32.add
set_local $l1
f64.const 0x1p+0 (;=1;)
set_local $l2
i32.const -1
set_local $p0
f64.const 0x1p+0 (;=1;)
set_local $l3
loop $L2
get_local $l3
get_local $p0
f64.convert_s/i32
get_local $l2
get_local $l2
f64.add
f64.const 0x1p+0 (;=1;)
f64.add
f64.div
f64.add
set_local $l3
get_local $l2
f64.const 0x1p+0 (;=1;)
f64.add
set_local $l2
i32.const 0
get_local $p0
i32.sub
set_local $p0
get_local $l1
i32.const -1
i32.add
tee_local $l1
br_if $L2
end
end
get_local $l3
f64.const 0x1p+2 (;=4;)
f64.mul
f64.const 0x1.1c37937e08p+53 (;=1e+16;)
f64.mul
tee_local $l2
f64.const 0x1p+64 (;=1.84467e+19;)
f64.lt
get_local $l2
f64.const 0x0p+0 (;=0;)
f64.ge
i32.and
br_if $B0
i64.const 0
set_local $l0
end
get_local $l0
return
end
get_local $l2
i64.trunc_u/f64)
(table $T0 1 1 anyfunc)
(memory $memory (export "memory") 2)
(global $g0 (mut i32) (i32.const 66560))
(global $__heap_base (export "__heap_base") i32 (i32.const 66560))
(global $__data_end (export "__data_end") i32 (i32.const 1024))) |
|
Agreed to make this use single precision. |
|
Here's the single-precision verison: (module
(type $t0 (func))
(type $t1 (func (param i32) (result i64)))
(func $__wasm_call_ctors (type $t0))
(func $taylor_pi (export "taylor_pi") (type $t1) (param $p0 i32) (result i64)
(local $l0 i64) (local $l1 i32) (local $l2 i32) (local $l3 f32) (local $l4 f32)
i64.const 40000001090256896
set_local $l0
block $B0
block $B1
get_local $p0
i32.const 2
i32.lt_u
br_if $B1
i32.const -1
set_local $l1
i32.const 1
set_local $l2
f32.const 0x1p+0 (;=1;)
set_local $l3
loop $L2
get_local $l3
get_local $l1
f32.convert_s/i32
get_local $l2
f32.convert_u/i32
tee_local $l4
get_local $l4
f32.add
f32.const 0x1p+0 (;=1;)
f32.add
f32.div
f32.add
set_local $l3
i32.const 0
get_local $l1
i32.sub
set_local $l1
get_local $p0
get_local $l2
i32.const 1
i32.add
tee_local $l2
i32.ne
br_if $L2
end
get_local $l3
f32.const 0x1p+2 (;=4;)
f32.mul
f32.const 0x1.1c3794p+53 (;=1e+16;)
f32.mul
tee_local $l3
f32.const 0x1p+64 (;=1.84467e+19;)
f32.lt
get_local $l3
f32.const 0x0p+0 (;=0;)
f32.ge
i32.and
br_if $B0
i64.const 0
set_local $l0
end
get_local $l0
return
end
get_local $l3
i64.trunc_u/f32)
(table $T0 1 1 anyfunc)
(memory $memory (export "memory") 2)
(global $g0 (mut i32) (i32.const 66560))
(global $__heap_base (export "__heap_base") i32 (i32.const 66560))
(global $__data_end (export "__data_end") i32 (i32.const 1024))) |
chfast
approved these changes
Aug 23, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Depends on #474.
This benchmark uses only f64 instructions: const, add, div, mul, lt, ge. And two conversions:
f64.convert_s/i32andi64.trunc_u/f64.