1 // Auto-generated file. Do not edit!
2 // Template: src/f32-ibilinear/scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2019 Google LLC
6 //
7 // This source code is licensed under the BSD-style license found in the
8 // LICENSE file in the root directory of this source tree.
9
10 #include <assert.h>
11
12 #include <xnnpack/ibilinear.h>
13
14
xnn_f32_ibilinear_ukernel__scalar_c4(size_t output_pixels,size_t channels,const float ** restrict input,size_t input_offset,const float * restrict weights,float * restrict output,size_t output_increment)15 void xnn_f32_ibilinear_ukernel__scalar_c4(
16 size_t output_pixels,
17 size_t channels,
18 const float**restrict input,
19 size_t input_offset,
20 const float*restrict weights,
21 float*restrict output,
22 size_t output_increment)
23 {
24 assert(output_pixels != 0);
25 assert(channels != 0);
26 assert(channels % sizeof(float) == 0);
27
28 do {
29 const float* i0 = (const float*) ((uintptr_t) input[0] + input_offset);
30 const float* i1 = (const float*) ((uintptr_t) input[1] + input_offset);
31 const float* i2 = (const float*) ((uintptr_t) input[2] + input_offset);
32 const float* i3 = (const float*) ((uintptr_t) input[3] + input_offset);
33 input += 4;
34
35 const float valphah = weights[0];
36 const float valphav = weights[1];
37 weights += 2;
38
39 size_t c = channels;
40 for (; c >= 4 * sizeof(float); c -= 4 * sizeof(float)) {
41 const float vtl0 = i0[0];
42 const float vtr0 = i1[0];
43 const float vbl0 = i2[0];
44 const float vbr0 = i3[0];
45 const float vtl1 = i0[1];
46 const float vtr1 = i1[1];
47 const float vbl1 = i2[1];
48 const float vbr1 = i3[1];
49 const float vtl2 = i0[2];
50 const float vtr2 = i1[2];
51 const float vbl2 = i2[2];
52 const float vbr2 = i3[2];
53 const float vtl3 = i0[3];
54 const float vtr3 = i1[3];
55 const float vbl3 = i2[3];
56 const float vbr3 = i3[3];
57 i0 += 4;
58 i1 += 4;
59 i2 += 4;
60 i3 += 4;
61
62 const float vtd0 = vtr0 - vtl0;
63 const float vbd0 = vbr0 - vbl0;
64 const float vtd1 = vtr1 - vtl1;
65 const float vbd1 = vbr1 - vbl1;
66 const float vtd2 = vtr2 - vtl2;
67 const float vbd2 = vbr2 - vbl2;
68 const float vtd3 = vtr3 - vtl3;
69 const float vbd3 = vbr3 - vbl3;
70
71 const float vt0 = vtl0 + vtd0 * valphah;
72 const float vb0 = vbl0 + vbd0 * valphah;
73 const float vt1 = vtl1 + vtd1 * valphah;
74 const float vb1 = vbl1 + vbd1 * valphah;
75 const float vt2 = vtl2 + vtd2 * valphah;
76 const float vb2 = vbl2 + vbd2 * valphah;
77 const float vt3 = vtl3 + vtd3 * valphah;
78 const float vb3 = vbl3 + vbd3 * valphah;
79
80 const float vd0 = vb0 - vt0;
81 const float vd1 = vb1 - vt1;
82 const float vd2 = vb2 - vt2;
83 const float vd3 = vb3 - vt3;
84
85 const float vo0 = vt0 + vd0 * valphav;
86 const float vo1 = vt1 + vd1 * valphav;
87 const float vo2 = vt2 + vd2 * valphav;
88 const float vo3 = vt3 + vd3 * valphav;
89
90 output[0] = vo0;
91 output[1] = vo1;
92 output[2] = vo2;
93 output[3] = vo3;
94 output += 4;
95 }
96 for (; c >= sizeof(float); c -= sizeof(float)) {
97 const float vtl = *i0++;
98 const float vtr = *i1++;
99 const float vbl = *i2++;
100 const float vbr = *i3++;
101
102 const float vtd = vtr - vtl;
103 const float vbd = vbr - vbl;
104
105 const float vt = vtl + vtd * valphah;
106 const float vb = vbl + vbd * valphah;
107
108 const float vd = vb - vt;
109
110 const float vo = vt + vd * valphav;
111
112 *output++ = vo;
113 }
114
115 output = (float*) ((uintptr_t) output + output_increment);
116 } while (--output_pixels != 0);
117 }
118