Skip to content

Commit

Permalink
benchmarks 0
Browse files Browse the repository at this point in the history
  • Loading branch information
@DenisYaroshevskiy
DenisYaroshevskiy committed Sep 10, 2023
1 parent 287a57a commit 4f4a3c6
Showing 1 changed file with 83 additions and 12 deletions.
95 changes: 83 additions & 12 deletions dyaroshev_presentations/conference_talks/advanced_simd_algorithms_in_pictures.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,9 +13,9 @@
<link rel="stylesheet" href="../../dist/theme/moon.css">

<script src="https://cdn.plot.ly/plotly-2.5.0.min.js"></script>
<script> const UNSQ_EVE_COMMIT = 'fb271e27790e043adea973ec77ad54cc6d7303b8'; </script>
<script> const UNSQ_EVE_COMMIT = '095883aedd2d5cd90cb59563be23e6b98e119714'; </script>
<script type="text/javascript"
src="https://cdn.jsdelivr.net/gh/DenisYaroshevskiy/unsq_eve@2cf6819dda923fc021629eda0cfba4113eb5f530/scripts/bench.js"></script>
src="https://cdn.jsdelivr.net/gh/DenisYaroshevskiy/unsq_eve@095883aedd2d5cd90cb59563be23e6b98e119714/scripts/bench.js"></script>

<script src="../scripts/helper_scripts.js"></script>

Expand All @@ -37,17 +37,68 @@ <h2> std::memcmp </h2>
<section id="advanced_simd_algorithms_in_pictures/memcmp">
<script> imagesSlideShow("advanced_simd_algorithms_in_pictures/memcmp", 5); </script>
</section>
<section>
<h2> PLACEHOLDER: std::memcmp is faster >4x std::mismatch(uint64_t)</h2>
<section id='int64 vs memcmp'>
<script>
addBenchmarkForParameters('int64 vs memcmp',
'Memcmp results, x86',
{
name: 'mismatch',
size: 10000,
algorithm: 'selection',
type: 'char',
time: 'y',
padding: 'min',
group: 'avx2+bmi',
},
['std::mismatch', 'use_int64s', 'std::memcmp'],
[]);
</script>
</section>
<section>
<h2> eve library </h2>
<ul>
<li class="fragment"> <a href="https://github.com/jfalcou/eve"> github </a></li>
<li class="fragment"> <a href="https://github.com/jfalcou/eve"> github
</a></li>
<li class="fragment"> Joel Falcou, Jean-Thierry Lapresté </li>
<li class="fragment"> eve::algo::mismatch </li>
</ul>
</section>
<section>
<section id='eve::mistamtch,x86,0'>
<script>
addBenchmarkForParameters('eve::mistamtch,x86,0',
'Memcmp results, x86',
{
name: 'mismatch',
size: 10000,
algorithm: 'selection',
type: 'char',
time: 'y',
padding: 'min',
group: 'avx2+bmi',
},
['std::mismatch', 'use_int64s', 'std::memcmp', 'eve::algo::mismatch'],
offByDefault = ['std::mismatch']);
</script>
</section>
<section id='eve::mistamtch,apple_m1,0'>
<script>
addBenchmarkForParameters('eve::mistamtch,apple_m1,0',
'Memcmp results, M1',
{
name: 'mismatch',
size: 10000,
algorithm: 'selection',
type: 'char',
time: 'y',
padding: 'min',
group: 'apple_m1',
},
['std::mismatch', 'use_int64s', 'std::memcmp', 'eve::algo::mismatch'],
offByDefault = ['std::mismatch']);
</script>
</section>
</section>
<section>
<h2>Vector Processor Extensions</h2>
<ul>
Expand All @@ -66,8 +117,22 @@ <h2>Vector Processor Extensions</h2>
<li>WASM</li>
</ul>
</section>
<section>
<h2> PLACEHOLDER: std::memcmp == eve/avx2 and 2x faster eve/sse2 </h2>
<section id='eve::mistamtch,arch'>
<script>
addBenchmarkForParameters('eve::mistamtch,arch',
'Marching',
{
name: 'mismatch',
size: 10000,
algorithm: 'selection',
type: 'char',
time: 'y',
padding: 'min',
group: 'selection',
},
['std::memcmp/avx2+bmi', 'eve::algo::mismatch/avx2+bmi', 'eve::algo::mismatch/sse2'],
offByDefault = ['std::mismatch']);
</script>
</section>
<section>
<h2> What's inside mismatch? </h2>
Expand All @@ -77,7 +142,8 @@ <h2> What's inside mismatch? </h2>
</section>
<section>
<h3> mismatch from scratch </h3>
<iframe width="1600px" height="550px" src="https://godbolt.org/e#g:!((g:!((g:!((h:codeEditor,i:(filename:'1',fontScale:18,fontUsePx:'0',j:1,lang:c%2B%2B,selection:(endColumn:1,endLineNumber:3,positionColumn:1,positionLineNumber:3,selectionStartColumn:1,selectionStartLineNumber:3,startColumn:1,startLineNumber:3),source:'%23include+%3Ceve/module/core.hpp%3E%0A%0Abool+core_of_mmatch(const+std::uint8_t*%26+a,+const+std::uint8_t*%26+b)%0A%7B%0A++using+wide+%3D+eve::wide%3Cstd::uint8_t%3E%3B%0A%0A++wide+wide_a+%3D+eve::load(a)%3B%0A++wide+wide_b+%3D+eve::load(b)%3B%0A%0A++eve::logical%3Cwide%3E+test+%3D+wide_a+!!%3D+wide_b%3B%0A++std::optional%3Cstd::ptrdiff_t%3E+match+%3D+eve::first_true(test)%3B%0A%0A++if+(!!match)+%7B%0A++++a+%2B%3D+wide::size()%3B%0A++++b+%2B%3D+wide::size()%3B%0A++++return+false%3B%0A++%7D%0A%0A++a+%2B%3D+*match%3B%0A++b+%2B%3D+*match%3B%0A++return+true%3B%0A%7D'),l:'5',n:'0',o:'C%2B%2B+source+%231',t:'0')),k:100,l:'4',m:82.3650653059978,n:'0',o:'',s:0,t:'0'),(g:!((h:compiler,i:(compiler:clang1500,deviceViewOpen:'1',filters:(b:'0',binary:'1',binaryObject:'1',commentOnly:'0',debugCalls:'1',demangle:'0',directives:'0',execute:'1',intel:'0',libraryCode:'0',trim:'1'),flagsViewOpen:'1',fontScale:18,fontUsePx:'0',j:1,lang:c%2B%2B,libs:!((name:eve,ver:v20230215)),options:'-mavx2+-DNDEBUG+--std%3Dc%2B%2B20+-O3',overrides:!(),selection:(endColumn:1,endLineNumber:1,positionColumn:1,positionLineNumber:1,selectionStartColumn:1,selectionStartLineNumber:1,startColumn:1,startLineNumber:1),source:1),l:'5',n:'0',o:'+x86-64+clang+15.0.0+(Editor+%231)',t:'0')),header:(),l:'4',m:17.6349346940022,n:'0',o:'',s:0,t:'0')),l:'3',n:'0',o:'',t:'0')),version:4"></iframe>
<iframe width="1600px" height="550px"
src="https://godbolt.org/e#g:!((g:!((g:!((h:codeEditor,i:(filename:'1',fontScale:18,fontUsePx:'0',j:1,lang:c%2B%2B,selection:(endColumn:1,endLineNumber:3,positionColumn:1,positionLineNumber:3,selectionStartColumn:1,selectionStartLineNumber:3,startColumn:1,startLineNumber:3),source:'%23include+%3Ceve/module/core.hpp%3E%0A%0Abool+core_of_mmatch(const+std::uint8_t*%26+a,+const+std::uint8_t*%26+b)%0A%7B%0A++using+wide+%3D+eve::wide%3Cstd::uint8_t%3E%3B%0A%0A++wide+wide_a+%3D+eve::load(a)%3B%0A++wide+wide_b+%3D+eve::load(b)%3B%0A%0A++eve::logical%3Cwide%3E+test+%3D+wide_a+!!%3D+wide_b%3B%0A++std::optional%3Cstd::ptrdiff_t%3E+match+%3D+eve::first_true(test)%3B%0A%0A++if+(!!match)+%7B%0A++++a+%2B%3D+wide::size()%3B%0A++++b+%2B%3D+wide::size()%3B%0A++++return+false%3B%0A++%7D%0A%0A++a+%2B%3D+*match%3B%0A++b+%2B%3D+*match%3B%0A++return+true%3B%0A%7D'),l:'5',n:'0',o:'C%2B%2B+source+%231',t:'0')),k:100,l:'4',m:82.3650653059978,n:'0',o:'',s:0,t:'0'),(g:!((h:compiler,i:(compiler:clang1500,deviceViewOpen:'1',filters:(b:'0',binary:'1',binaryObject:'1',commentOnly:'0',debugCalls:'1',demangle:'0',directives:'0',execute:'1',intel:'0',libraryCode:'0',trim:'1'),flagsViewOpen:'1',fontScale:18,fontUsePx:'0',j:1,lang:c%2B%2B,libs:!((name:eve,ver:v20230215)),options:'-mavx2+-DNDEBUG+--std%3Dc%2B%2B20+-O3',overrides:!(),selection:(endColumn:1,endLineNumber:1,positionColumn:1,positionLineNumber:1,selectionStartColumn:1,selectionStartLineNumber:1,startColumn:1,startLineNumber:1),source:1),l:'5',n:'0',o:'+x86-64+clang+15.0.0+(Editor+%231)',t:'0')),header:(),l:'4',m:17.6349346940022,n:'0',o:'',s:0,t:'0')),l:'3',n:'0',o:'',t:'0')),version:4"></iframe>
</section>
<section id="advanced_simd_algorithms_in_pictures/first_true">
<script> imagesSlideShow("advanced_simd_algorithms_in_pictures/first_true", 6); </script>
Expand Down Expand Up @@ -118,7 +184,8 @@ <h2> compress_copy (pure simd) </h2>
<li class="fragment"> switch + shuffle (@Z boson) </li>
</ul>
</section>
<section id="advanced_simd_algorithms_in_pictures/compress_copy_simd_scalar">
<section
id="advanced_simd_algorithms_in_pictures/compress_copy_simd_scalar">
<script> imagesSlideShow("advanced_simd_algorithms_in_pictures/compress_copy_simd_scalar", 4); </script>
</section>
<section>
Expand All @@ -144,17 +211,21 @@ <h2> compress_copy (simd/scalar) </h2>
<section>
<h2> set_intersection </h2>
</section>
<section id="advanced_simd_algorithms_in_pictures/set_intersection_simd_scalar">
<section
id="advanced_simd_algorithms_in_pictures/set_intersection_simd_scalar">
<script> imagesSlideShow("advanced_simd_algorithms_in_pictures/set_intersection_simd_scalar", 9); </script>
</section>
<section>
<h2> set intersection (simd/simd) </h2>
<ul>
<li class="fragment"> <a href="https://arxiv.org/abs/2112.06342">Faster-Than-Native Alternatives for x86 VP2INTERSECT Instructions</a></li>
<li class="fragment"> <a
href="https://arxiv.org/abs/2112.06342">Faster-Than-Native
Alternatives for x86 VP2INTERSECT Instructions</a></li>
<li class="fragment"> Guillermo Diez-Canas </li>
</ul>
</section>
<section id="advanced_simd_algorithms_in_pictures/set_intersection_simd_simd">
<section
id="advanced_simd_algorithms_in_pictures/set_intersection_simd_simd">
<script> imagesSlideShow("advanced_simd_algorithms_in_pictures/set_intersection_simd_simd", 8); </script>
</section>
<section id="advanced_simd_algorithms_in_pictures/has_equal_in">
Expand Down

0 comments on commit 4f4a3c6

Please sign in to comment.