feat(simd): add clampn4_f32, sum4_f32

packages/simd/README.md

@@ -23,7 +23,7 @@ This project is part of the
 ## About
 
 [WebAssembly SIMD](https://github.com/WebAssembly/simd) vector
-operations for batch processing, written in
+operations for array/batch processing, written in
 [AssemblyScript](https://docs.assemblyscript.org/).
 
 ## Available functions
@@ -36,6 +36,7 @@ for sources:
 - `add4_f32`
 - `addn4_f32`
 - `clamp4_f32`
+- `clampn4_f32`
 - `div4_f32` (*)
 - `divn4_f32` (*)
 - `dot2_f32_aos` (2x vec2 per iteration)
@@ -57,6 +58,7 @@ for sources:
 - `sqrt4_f32` (*)
 - `sub4_f32`
 - `subn4_f32`
+- `sum4_f32`
 
 (*) Missing native implementation, waiting on...
 

packages/simd/assembly/clamp.ts

@@ -47,3 +47,25 @@ export function clamp4_f32(
     }
     return res;
 }
+
+export function clampn4_f32(
+    out: usize,
+    a: usize,
+    b: f32,
+    c: f32,
+    num: usize,
+    so: usize,
+    sa: usize
+): usize {
+    const res = out;
+    so <<= 2;
+    sa <<= 2;
+    const vmin = f32x4.splat(b);
+    const vmax = f32x4.splat(c);
+    for (; num-- > 0; ) {
+        v128.store(out, f32x4.min(f32x4.max(v128.load(a), vmin), vmax));
+        out += so;
+        a += sa;
+    }
+    return res;
+}

packages/simd/assembly/index.ts

@@ -25,3 +25,4 @@ export * from "./normalize";
 
 export * from "./sub";
 export * from "./subn";
+export * from "./sum";

packages/simd/assembly/sum.ts

@@ -0,0 +1,10 @@
+export function sum4_f32(a: usize, num: usize, sa: usize): f64 {
+    sa <<= 2;
+    let acc = f32x4.splat(0);
+    for (; num-- > 0; ) {
+        acc = f32x4.add(acc, v128.load(a));
+        a += sa;
+    }
+    acc = f32x4.add(acc, v128.shuffle<f32>(acc, acc, 2, 3, 0, 1));
+    return f32x4.extract_lane(acc, 0) + f32x4.extract_lane(acc, 1);
+}

packages/simd/src/api.ts

@@ -11,6 +11,9 @@ export interface SIMD {
     // prettier-ignore
     clamp4_f32(out: number, a: number, b: number, c: number, num: number, so: number, sa: number, sb: number, sc: number): number;
 
+    // prettier-ignore
+    clampn4_f32(out: number, a: number, b: number, c: number, num: number, so: number, sa: number): number;
+
     // prettier-ignore
     div4_f32(out: number, a: number, b: number, num: number, so: number, sa: number, sb: number): number;
 
@@ -159,6 +162,8 @@ export interface SIMD {
     // prettier-ignore
     subn4_f32(out: number, a: number, n: number, num: number, so: number, sa: number): number;
 
+    sum4_f32(a: number, num: number, sa: number): number;
+
     /**
      * WASM memory instance given to `init()`.
      */