feat(simd): enable new ops supported in node 14/V8 8.3

- previously unsupported in V8 impl (f32x4.div & f32x4.sqrt ops): - div4_f32 - divn4_f32 - sqrt4_f32 - invsqrt4_f32
thi-ng · May 5, 2020 · 5c46468 · 5c46468
1 parent 744a32b
commit 5c46468
Show file tree

Hide file tree

Showing 3 changed files with 28 additions and 15 deletions.
diff --git a/packages/simd/assembly/index.ts b/packages/simd/assembly/index.ts
@@ -4,8 +4,8 @@ export * from "./addn";
 export * from "./clamp";
 
 // TODO waiting for native impl
-// export * from "./div";
-// export * from "./divn";
+export * from "./div";
+export * from "./divn";
 
 export * from "./dot";
 export * from "./madd";
@@ -25,7 +25,7 @@ export * from "./neg";
 export * from "./normalize";
 
 // TODO waiting for native impl
-// export * from "./sqrt";
+export * from "./sqrt";
 
 export * from "./sub";
 export * from "./subn";

diff --git a/packages/simd/src/api.ts b/packages/simd/src/api.ts
@@ -73,9 +73,7 @@ export interface SIMD {
     dot4_f32_soa(out: number, a: number, b: number, num: number, sa: number, sb: number): number;
 
     /**
-     * FIXME waiting for native impl of SIMD instr
-     *
-     * {@link SIMD.sqrt4_f32}
+     * Also see {@link SIMD.sqrt4_f32}
      *
      * @param out -
      * @param a -
@@ -163,9 +161,7 @@ export interface SIMD {
     normalize4_f32_aos(out: number, a: number, num: number, norm: number, so: number, sa: number): number;
 
     /**
-     * FIXME waiting for native impl of SIMD instr
-     *
-     * {@link SIMD.invsqrt4_f32}
+     * Also see {@link SIMD.invsqrt4_f32}
      *
      * @param out -
      * @param a -

diff --git a/packages/simd/test/index.ts b/packages/simd/test/index.ts
@@ -51,11 +51,11 @@ assertEqual(res_f32(1024, 8), [10, 40, 90, 160, 250, 360, 490, 640]);
 simd.muln4_f32(1024, 0, 10, 2, 4, 4);
 assertEqual(res_f32(1024, 8), [10, 20, 30, 40, 50, 60, 70, 80]);
 // div4_f32
-// simd.div4_f32(1024, 0, 48, 2, 4, 4, 4);
-// assertEqualDelta(res_f32(1024, 8), [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1]);
+simd.div4_f32(1024, 0, 48, 2, 4, 4, 4);
+assertEqualDelta(res_f32(1024, 8), [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1]);
 // divn4_f32
-// simd.divn4_f32(1024, 0, 10, 2, 4, 4);
-// assertEqualDelta(res_f32(1024, 8), [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]);
+simd.divn4_f32(1024, 0, 10, 2, 4, 4);
+assertEqualDelta(res_f32(1024, 8), [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]);
 
 // abs4_f32
 simd.f32.set([-1, 2, -3, 4, 5, -6, 7, -8]);
@@ -178,21 +178,38 @@ assertEqualDelta(res_f32(1024, 4), [
     1 * 1 + 2 * 2,
     10 * 10 + 20 * 20,
     100 * 100 + 200 * 200,
-    100 * 100 + 200 * 200
+    100 * 100 + 200 * 200,
 ]);
 simd.mag2_f32_aos(1024, 0, 4);
 assertEqualDelta(res_f32(1024, 4), [
     Math.sqrt(5),
     Math.sqrt(500),
     Math.sqrt(50000),
-    Math.sqrt(50000)
+    Math.sqrt(50000),
 ]);
 
 simd.magsq4_f32_aos(1024, 0, 2, 1, 4);
 assertEqualDelta(res_f32(1024, 2), [505, 100000]);
 simd.mag4_f32_aos(1024, 0, 2, 1, 4);
 assertEqualDelta(res_f32(1024, 2), [Math.sqrt(505), Math.sqrt(100000)]);
 
+// sqrt4
+simd.f32.set([1, 2, 9, 16, 25, 36, 49, 64]);
+simd.sqrt4_f32(1024, 0, 2, 4, 4);
+assertEqualDelta(res_f32(1024, 8), [1, Math.SQRT2, 3, 4, 5, 6, 7, 8]);
+// invsqrt4
+simd.invsqrt4_f32(1024, 0, 2, 4, 4);
+assertEqualDelta(res_f32(1024, 8), [
+    1,
+    Math.SQRT1_2,
+    1 / 3,
+    1 / 4,
+    1 / 5,
+    1 / 6,
+    1 / 7,
+    1 / 8,
+]);
+
 // mix4_f32
 // mixn4_f32
 // prettier-ignore