• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2020 Google LLC
2 //
3 // This source code is licensed under the BSD-style license found in the
4 // LICENSE file in the root directory of this source tree.
5 
6 #include <assert.h>
7 
8 #include <wasm_simd128.h>
9 
10 #include <xnnpack/zip.h>
11 
12 
xnn_x32_zip_x3_ukernel__wasmsimd(size_t n,const uint32_t * input,uint32_t * output)13 void xnn_x32_zip_x3_ukernel__wasmsimd(
14     size_t n,
15     const uint32_t* input,
16     uint32_t* output)
17 {
18   assert(n != 0);
19   assert(n % sizeof(uint32_t) == 0);
20 
21   const float* x = (const float*) input;
22   const float* y = (const float*) ((uintptr_t) x + n);
23   const float* z = (const float*) ((uintptr_t) y + n);
24   float* o = (float*) output;
25 
26   while (n >= 4 * sizeof(uint32_t)) {
27     // vx = ( x3, x2, x1, x0 )
28     const v128_t vx = wasm_v128_load(x);
29     x += 4;
30     // vy = ( y3, y2, y1, y0 )
31     const v128_t vy = wasm_v128_load(y);
32     y += 4;
33     // vz = ( z3, z2, z1, z0 )
34     const v128_t vz = wasm_v128_load(z);
35     z += 4;
36 
37     // vxy = ( y2, y0, x2, x0 )
38     const v128_t vxy = wasm_v32x4_shuffle(vx, vy, 0, 2, 4, 6);
39     // vyz = ( z3, z1, y3, y1 )
40     const v128_t vyz = wasm_v32x4_shuffle(vy, vz, 1, 3, 5, 7);
41     // vzx = ( x3, x1, z2, z0 )
42     const v128_t vzx = wasm_v32x4_shuffle(vz, vx, 0, 2, 5, 7);
43 
44     // vxyz0 = ( x1, z0, y0, x0 )
45     const v128_t vxyz0 = wasm_v32x4_shuffle(vxy, vzx, 0, 2, 4, 6);
46     // vxyz1 = ( y2, x2, z1, y1 )
47     const v128_t vxyz1 = wasm_v32x4_shuffle(vyz, vxy, 0, 2, 5, 7);
48     // vxyz2 = ( z3, y3, x3, z2 )
49     const v128_t vxyz2 = wasm_v32x4_shuffle(vzx, vyz, 1, 3, 5, 7);
50 
51     wasm_v128_store(o, vxyz0);
52     wasm_v128_store(o + 4, vxyz1);
53     wasm_v128_store(o + 8, vxyz2);
54     o += 12;
55     n -= 4 * sizeof(uint32_t);
56   }
57   if XNN_UNLIKELY(n != 0) {
58     do {
59       const float vx = *x++;
60       const float vy = *y++;
61       const float vz = *z++;
62       o[0] = vx;
63       o[1] = vy;
64       o[2] = vz;
65       o += 3;
66       n -= sizeof(uint32_t);
67     } while (n != 0);
68   }
69 }
70