xref: /aosp_15_r20/external/XNNPACK/src/cs16-vsquareabs/gen/neon-mlal-ld128-x4.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Auto-generated file. Do not edit!
2 //   Template: src/cs16-vsquareabs/neon.c.in
3 //   Generator: tools/xngen
4 //
5 // Copyright 2022 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 #include <stddef.h>
12 #include <stdint.h>
13 
14 #include <arm_neon.h>
15 
16 #include <xnnpack/vsquareabs.h>
17 
18 
xnn_cs16_vsquareabs_ukernel__neon_mlal_ld128_x4(size_t batch,const int16_t * input,uint32_t * output)19 void xnn_cs16_vsquareabs_ukernel__neon_mlal_ld128_x4(
20     size_t batch,
21     const int16_t* input,
22     uint32_t* output) {
23 
24   assert(batch != 0);
25   assert(input != NULL);
26   assert(output != NULL);
27 
28 
29   // Remainder of full vectors
30   for (; batch >= 4; batch -= 4) {
31     const int16x4x2_t vi = vld2_s16(input); input += 8;
32 
33     int32x4_t vacc = vmull_s16(vi.val[0], vi.val[0]);
34 
35     vacc = vmlal_s16(vacc, vi.val[1], vi.val[1]);
36 
37     vst1q_u32(output, vreinterpretq_u32_s32(vacc)); output += 4;
38   }
39 
40   // Remainder of 1 to 3 elements
41   if XNN_UNLIKELY(batch != 0) {
42     const int16x4x2_t vi = vld2_s16(input);
43 
44     int32x4_t vacc = vmull_s16(vi.val[0], vi.val[0]);
45     vacc = vmlal_s16(vacc, vi.val[1], vi.val[1]);
46 
47     uint32x2_t vacc_lo = vreinterpret_u32_s32(vget_low_s32(vacc));
48     if (batch & 2) {
49       vst1_u32(output, vacc_lo); output += 2;
50       vacc_lo = vreinterpret_u32_s32(vget_high_s32(vacc));
51     }
52     if (batch & 1) {
53       vst1_lane_u32(output, vacc_lo, 0);
54     }
55   }
56 }
57