[WTF] Add memchr16 on ARM64

https://bugs.webkit.org/show_bug.cgi?id=242568 Reviewed by Darin Adler. This change adds SIMD implementation of memchr16. We use ARM64 NEON intrinsics to implement faster memchr16, and leverage it in WTF::find and TypedArray indexOf/includes. ToT Patched u16-string-index-of-1001-mid 0.6371+-0.0201 ^ 0.4996+-0.0310 ^ definitely 1.2753x faster u16-string-index-of-101-beg 0.4603+-0.0461 0.4518+-0.0241 might be 1.0186x faster u16-string-index-of-11-end 0.4592+-0.0350 ? 0.4787+-0.0359 ? might be 1.0424x slower u16-string-index-of-100001-end 33.0156+-0.6301 ^ 6.4448+-0.0449 ^ definitely 5.1229x faster uint16-array-index-of-large 665.2918+-2.5632 ^ 112.1221+-1.7009 ^ definitely 5.9336x faster u16-string-index-of-10000001-end 3261.2755+-5.9943 ^ 622.5246+-1.2345 ^ definitely 5.2388x faster u16-string-index-of-101-404 0.4759+-0.0301 ? 0.5007+-0.0583 ? might be 1.0523x slower u16-string-index-of-10001-mid 2.1295+-0.0140 ^ 0.8355+-0.0185 ^ definitely 2.5487x faster u16-string-index-of-1000001-mid 168.4613+-3.2181 ^ 35.4271+-0.0338 ^ definitely 4.7551x faster uint16-array-index-of-small 0.4010+-0.0118 0.3834+-0.0151 might be 1.0461x faster u16-string-index-of-101-end 0.4986+-0.0699 0.4568+-0.0184 might be 1.0914x faster u16-string-index-of-11-beg 0.4497+-0.0382 0.4433+-0.0378 might be 1.0144x faster u16-string-index-of-100001-beg 1.4841+-0.0359 ? 1.5071+-0.0142 ? might be 1.0155x slower u16-string-index-of-10000001-404 3257.2980+-4.5640 ^ 625.2343+-3.0194 ^ definitely 5.2097x faster int16-array-index-of-large 665.0809+-3.2257 ^ 112.5059+-2.9796 ^ definitely 5.9115x faster u16-string-index-of-100001-404 33.4299+-1.1079 ^ 6.5370+-0.2573 ^ definitely 5.1140x faster u16-string-index-of-11-404 0.4853+-0.0453 0.4795+-0.0406 might be 1.0121x faster int16-array-index-of-small 0.6745+-0.4688 0.3655+-0.0104 might be 1.8454x faster u16-string-index-of-10000001-beg 112.7112+-0.9075 ! 116.7191+-2.8648 ! definitely 1.0356x slower u16-string-index-of-10000001-mid 1721.3761+-21.7927 ^ 377.8264+-1.9241 ^ definitely 4.5560x faster int16-array-index-of-medium 13.2849+-0.0279 ^ 2.5942+-0.0163 ^ definitely 5.1211x faster uint16-array-index-of-medium 13.4077+-0.1505 ^ 2.5650+-0.0501 ^ definitely 5.2272x faster u16-string-index-of-10001-end 3.6959+-0.0262 ^ 1.0838+-0.0770 ^ definitely 3.4100x faster u16-string-index-of-1000001-end 328.2864+-7.9285 ^ 59.9328+-0.3556 ^ definitely 5.4776x faster u16-string-index-of-1001-end 0.8068+-0.0256 ^ 0.5374+-0.0365 ^ definitely 1.5014x faster u16-string-index-of-11-mid 0.4565+-0.0454 0.4553+-0.0315 u16-string-index-of-100001-mid 17.3082+-0.4797 ^ 3.9760+-0.0264 ^ definitely 4.3532x faster u16-string-index-of-10001-beg 0.5774+-0.0328 0.5710+-0.0408 might be 1.0113x faster u16-string-index-of-1000001-beg 10.9870+-0.3981 ? 11.0094+-0.0433 ? u16-string-index-of-1001-404 0.8054+-0.0332 ^ 0.5104+-0.0227 ^ definitely 1.5778x faster u16-string-index-of-101-mid 0.4739+-0.0377 0.4641+-0.0256 might be 1.0212x faster u16-string-index-of-1000001-404 326.1748+-5.7428 ^ 60.2302+-1.2138 ^ definitely 5.4155x faster u16-string-index-of-10001-404 3.6958+-0.0159 ^ 1.0830+-0.0501 ^ definitely 3.4125x faster u16-string-index-of-1001-beg 0.4613+-0.0361 ? 0.4815+-0.0357 ? might be 1.0438x slower <geometric> 7.7050+-0.0885 ^ 3.5582+-0.0275 ^ definitely 2.1654x faster * JSTests/microbenchmarks/int16-array-index-of-large.js: Added. (test): * JSTests/microbenchmarks/int16-array-index-of-medium.js: Added. (test): * JSTests/microbenchmarks/int16-array-index-of-small.js: Added. (test): * JSTests/microbenchmarks/u16-string-index-of-10000001-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10000001-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10000001-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10000001-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1000001-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1000001-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1000001-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1000001-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-100001-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-100001-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-100001-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-100001-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10001-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10001-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10001-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-10001-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1001-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1001-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1001-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-1001-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-101-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-101-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-101-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-101-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-11-404.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-11-beg.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-11-end.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/u16-string-index-of-11-mid.js: Added. (bench): (target): (forRepeatCount): (all): * JSTests/microbenchmarks/uint16-array-index-of-large.js: Added. (test): * JSTests/microbenchmarks/uint16-array-index-of-medium.js: Added. (test): * JSTests/microbenchmarks/uint16-array-index-of-small.js: Added. (test): * Source/JavaScriptCore/runtime/JSGenericTypedArrayViewPrototypeFunctions.h: (JSC::genericTypedArrayViewProtoFuncIncludes): (JSC::genericTypedArrayViewProtoFuncIndexOf): * Source/WTF/WTF.xcodeproj/project.pbxproj: * Source/WTF/wtf/CMakeLists.txt: * Source/WTF/wtf/text/StringCommon.cpp: Added. (WTF::memchr16): * Source/WTF/wtf/text/StringCommon.h: (WTF::find): Canonical link: https://commits.webkit.org/252544@main
WebKit · Jul 17, 2022 · 3942acb · 3942acb
1 parent 6545a6f
commit 3942acb
Show file tree

Hide file tree

Showing 40 changed files with 1,975 additions and 13 deletions.
diff --git a/JSTests/microbenchmarks/int16-array-index-of-large.js b/JSTests/microbenchmarks/int16-array-index-of-large.js
@@ -0,0 +1,14 @@
+let size = 20 << 20;
+let array = new Int16Array(size);
+for (let i = 0; i < size; ++i)
+    array[i] = 1;
+array[size - 1] = 42;
+
+function test(array) {
+    let result = 0;
+    for (let i = 0; i < 1e2; ++i)
+        result += array.indexOf(42);
+    return result;
+}
+noInline(test);
+test(array);
diff --git a/JSTests/microbenchmarks/int16-array-index-of-medium.js b/JSTests/microbenchmarks/int16-array-index-of-medium.js
@@ -0,0 +1,14 @@
+let size = 4096;
+let array = new Int16Array(size);
+for (let i = 0; i < size; ++i)
+    array[i] = 1;
+array[size - 1] = 42;
+
+function test(array) {
+    let result = 0;
+    for (let i = 0; i < 1e4; ++i)
+        result += array.indexOf(42);
+    return result;
+}
+noInline(test);
+test(array);
diff --git a/JSTests/microbenchmarks/int16-array-index-of-small.js b/JSTests/microbenchmarks/int16-array-index-of-small.js
@@ -0,0 +1,14 @@
+let size = 16;
+let array = new Int16Array(size);
+for (let i = 0; i < size; ++i)
+    array[i] = 1;
+array[size - 1] = 42;
+
+function test(array) {
+    let result = 0;
+    for (let i = 0; i < 1e4; ++i)
+        result += array.indexOf(42);
+    return result;
+}
+noInline(test);
+test(array);
diff --git a/JSTests/microbenchmarks/u16-string-index-of-10000001-404.js b/JSTests/microbenchmarks/u16-string-index-of-10000001-404.js
@@ -0,0 +1,61 @@
+function bench(string, func)
+{
+    for (var i = 0; i < 1000; ++i)
+        func();
+}
+noInline(bench);
+
+function forRepeatCount(count, pos, utf16) {
+    var base = "lalalalala".repeat(count);
+    if (utf16) {
+        base += "ϧ"; // arbitrary utf-16
+    }
+
+    var input = base;
+    var label;
+    const charToFind = !utf16 ? "z" : String.fromCodePoint(0x0245);
+    switch (pos) {
+        case -1: {
+            input = charToFind + base;
+            label = `beg ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+        case 0: {
+            label = `mid ${utf16 ? "UChar" : "LChar"}`;
+            input =
+            base.substring(0, (base.length / 2) | 0) +
+            charToFind +
+            base.substring((base.length / 2) | 0);
+            break;
+        }
+        case 1: {
+            label = `end ${utf16 ? "UChar" : "LChar"}`;
+            input = base + charToFind;
+            break;
+        }
+            // not found
+        case 2: {
+            label = `404 ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+    }
+
+    // force it to not be a rope
+    input = input.split("").join("");
+
+    function target() {
+        input.indexOf(charToFind)
+    }
+    noInline(target);
+    return bench(
+        `<${label}> [${new Intl.NumberFormat()
+                .format(input.length)
+                .padStart("10,000,001".length)} chars] indexOf`,
+        target
+    );
+}
+
+function all(utf16) {
+    forRepeatCount(1000000, 2, !!utf16);
+}
+all(true);
diff --git a/JSTests/microbenchmarks/u16-string-index-of-10000001-beg.js b/JSTests/microbenchmarks/u16-string-index-of-10000001-beg.js
@@ -0,0 +1,61 @@
+function bench(string, func)
+{
+    for (var i = 0; i < 1000; ++i)
+        func();
+}
+noInline(bench);
+
+function forRepeatCount(count, pos, utf16) {
+    var base = "lalalalala".repeat(count);
+    if (utf16) {
+        base += "ϧ"; // arbitrary utf-16
+    }
+
+    var input = base;
+    var label;
+    const charToFind = !utf16 ? "z" : String.fromCodePoint(0x0245);
+    switch (pos) {
+        case -1: {
+            input = charToFind + base;
+            label = `beg ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+        case 0: {
+            label = `mid ${utf16 ? "UChar" : "LChar"}`;
+            input =
+            base.substring(0, (base.length / 2) | 0) +
+            charToFind +
+            base.substring((base.length / 2) | 0);
+            break;
+        }
+        case 1: {
+            label = `end ${utf16 ? "UChar" : "LChar"}`;
+            input = base + charToFind;
+            break;
+        }
+            // not found
+        case 2: {
+            label = `404 ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+    }
+
+    // force it to not be a rope
+    input = input.split("").join("");
+
+    function target() {
+        input.indexOf(charToFind)
+    }
+    noInline(target);
+    return bench(
+        `<${label}> [${new Intl.NumberFormat()
+                .format(input.length)
+                .padStart("10,000,001".length)} chars] indexOf`,
+        target
+    );
+}
+
+function all(utf16) {
+    forRepeatCount(1000000, -1, !!utf16);
+}
+all(true);
diff --git a/JSTests/microbenchmarks/u16-string-index-of-10000001-end.js b/JSTests/microbenchmarks/u16-string-index-of-10000001-end.js
@@ -0,0 +1,61 @@
+function bench(string, func)
+{
+    for (var i = 0; i < 1000; ++i)
+        func();
+}
+noInline(bench);
+
+function forRepeatCount(count, pos, utf16) {
+    var base = "lalalalala".repeat(count);
+    if (utf16) {
+        base += "ϧ"; // arbitrary utf-16
+    }
+
+    var input = base;
+    var label;
+    const charToFind = !utf16 ? "z" : String.fromCodePoint(0x0245);
+    switch (pos) {
+        case -1: {
+            input = charToFind + base;
+            label = `beg ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+        case 0: {
+            label = `mid ${utf16 ? "UChar" : "LChar"}`;
+            input =
+            base.substring(0, (base.length / 2) | 0) +
+            charToFind +
+            base.substring((base.length / 2) | 0);
+            break;
+        }
+        case 1: {
+            label = `end ${utf16 ? "UChar" : "LChar"}`;
+            input = base + charToFind;
+            break;
+        }
+            // not found
+        case 2: {
+            label = `404 ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+    }
+
+    // force it to not be a rope
+    input = input.split("").join("");
+
+    function target() {
+        input.indexOf(charToFind)
+    }
+    noInline(target);
+    return bench(
+        `<${label}> [${new Intl.NumberFormat()
+                .format(input.length)
+                .padStart("10,000,001".length)} chars] indexOf`,
+        target
+    );
+}
+
+function all(utf16) {
+    forRepeatCount(1000000, 1, !!utf16);
+}
+all(true);
diff --git a/JSTests/microbenchmarks/u16-string-index-of-10000001-mid.js b/JSTests/microbenchmarks/u16-string-index-of-10000001-mid.js
@@ -0,0 +1,61 @@
+function bench(string, func)
+{
+    for (var i = 0; i < 1000; ++i)
+        func();
+}
+noInline(bench);
+
+function forRepeatCount(count, pos, utf16) {
+    var base = "lalalalala".repeat(count);
+    if (utf16) {
+        base += "ϧ"; // arbitrary utf-16
+    }
+
+    var input = base;
+    var label;
+    const charToFind = !utf16 ? "z" : String.fromCodePoint(0x0245);
+    switch (pos) {
+        case -1: {
+            input = charToFind + base;
+            label = `beg ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+        case 0: {
+            label = `mid ${utf16 ? "UChar" : "LChar"}`;
+            input =
+            base.substring(0, (base.length / 2) | 0) +
+            charToFind +
+            base.substring((base.length / 2) | 0);
+            break;
+        }
+        case 1: {
+            label = `end ${utf16 ? "UChar" : "LChar"}`;
+            input = base + charToFind;
+            break;
+        }
+            // not found
+        case 2: {
+            label = `404 ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+    }
+
+    // force it to not be a rope
+    input = input.split("").join("");
+
+    function target() {
+        input.indexOf(charToFind)
+    }
+    noInline(target);
+    return bench(
+        `<${label}> [${new Intl.NumberFormat()
+                .format(input.length)
+                .padStart("10,000,001".length)} chars] indexOf`,
+        target
+    );
+}
+
+function all(utf16) {
+    forRepeatCount(1000000, 0, !!utf16);
+}
+all(true);
diff --git a/JSTests/microbenchmarks/u16-string-index-of-1000001-404.js b/JSTests/microbenchmarks/u16-string-index-of-1000001-404.js
@@ -0,0 +1,61 @@
+function bench(string, func)
+{
+    for (var i = 0; i < 1000; ++i)
+        func();
+}
+noInline(bench);
+
+function forRepeatCount(count, pos, utf16) {
+    var base = "lalalalala".repeat(count);
+    if (utf16) {
+        base += "ϧ"; // arbitrary utf-16
+    }
+
+    var input = base;
+    var label;
+    const charToFind = !utf16 ? "z" : String.fromCodePoint(0x0245);
+    switch (pos) {
+        case -1: {
+            input = charToFind + base;
+            label = `beg ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+        case 0: {
+            label = `mid ${utf16 ? "UChar" : "LChar"}`;
+            input =
+            base.substring(0, (base.length / 2) | 0) +
+            charToFind +
+            base.substring((base.length / 2) | 0);
+            break;
+        }
+        case 1: {
+            label = `end ${utf16 ? "UChar" : "LChar"}`;
+            input = base + charToFind;
+            break;
+        }
+            // not found
+        case 2: {
+            label = `404 ${utf16 ? "UChar" : "LChar"}`;
+            break;
+        }
+    }
+
+    // force it to not be a rope
+    input = input.split("").join("");
+
+    function target() {
+        input.indexOf(charToFind)
+    }
+    noInline(target);
+    return bench(
+        `<${label}> [${new Intl.NumberFormat()
+                .format(input.length)
+                .padStart("10,000,001".length)} chars] indexOf`,
+        target
+    );
+}
+
+function all(utf16) {
+    forRepeatCount(100000, 2, !!utf16);
+}
+all(true);