1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vscaleextexp/avx512f-p5-scalef.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 <immintrin.h>
13
14 #include <xnnpack/common.h>
15 #include <xnnpack/intrinsics-polyfill.h>
16 #include <xnnpack/vscaleextexp.h>
17
18
xnn_f32_vscaleextexp_ukernel__avx512f_p5_scalef_x16(size_t elements,const float * x,float * y,float scale_value,float scale_exp)19 void xnn_f32_vscaleextexp_ukernel__avx512f_p5_scalef_x16(
20 size_t elements,
21 const float* x,
22 float* y,
23 float scale_value,
24 float scale_exp)
25 {
26 assert(elements % sizeof(float) == 0);
27
28 const __m512 vlog2e = _mm512_set1_ps(0x1.715476p+0f);
29 const __m512 vminus_ln2_hi = _mm512_set1_ps(-0x1.62E43p-1f);
30 const __m512 vminus_ln2_lo = _mm512_set1_ps(0x1.05C61p-29f);
31
32 const __m512 vc0 = _mm512_set1_ps(1.0f);
33 const __m512 vc1 = _mm512_set1_ps(0x1.FFFFF6p-1f);
34 const __m512 vc2 = _mm512_set1_ps(0x1.FFFDC6p-2f);
35 const __m512 vc3 = _mm512_set1_ps(0x1.555A80p-3f);
36 const __m512 vc4 = _mm512_set1_ps(0x1.573A1Ap-5f);
37 const __m512 vc5 = _mm512_set1_ps(0x1.0F9F9Cp-7f);
38
39 const __m512 vscalev = _mm512_set1_ps(scale_value);
40 const __m512 vscalee = _mm512_set1_ps(scale_exp);
41
42 for (; elements >= 16 * sizeof(float); elements -= 16 * sizeof(float)) {
43 // Load 16 (1x16) inputs at a time.
44 const __m512 vx0 = _mm512_loadu_ps(x);
45 x += 16;
46
47 // Compute reduced argument elements := round(x / log(2)).
48 const __m512 vn0 = _mm512_roundscale_ps(_mm512_mul_ps(vx0, vlog2e), 0);
49
50 // Compute reduced argument t := x - elements * log(2).
51 // Use Cody-Waite range reduction method (note two constants to represent log(2)) to improve accuracy.
52 __m512 vt0 = _mm512_fmadd_ps(vn0, vminus_ln2_hi, vx0);
53
54 vt0 = _mm512_fmadd_ps(vn0, vminus_ln2_lo, vt0);
55
56 // Compute degree-5 polynomial approximation for exp(t) on [-log(2)/2, log(2)/2].
57 __m512 vp0 = _mm512_fmadd_ps(vc5, vt0, vc4);
58
59 vp0 = _mm512_fmadd_ps(vp0, vt0, vc3);
60
61 vp0 = _mm512_fmadd_ps(vp0, vt0, vc2);
62
63 vp0 = _mm512_fmadd_ps(vp0, vt0, vc1);
64
65 vp0 = _mm512_fmadd_ps(vp0, vt0, vc0);
66
67 // Multiply "extended" floating-point numbers in ("mantissa", "exponent") representation where
68 // - vnX is "exponent"
69 // - vpX is "mantissa"
70 //
71 // exp2(ae) * av * exp2(be) * bv =
72 // = exp2(ae + be) * (av * bv)
73 __m512 vf0 = _mm512_mul_ps(vp0, vscalev);
74
75 const __m512 ve0 = _mm512_add_ps(vn0, vscalee);
76
77 // Multiply "mantissa" by the exp2("exponent").
78 vf0 = _mm512_scalef_ps(vf0, ve0);
79
80 // Store 128 (8x16) results at a time.
81 _mm512_storeu_ps(y, vf0);
82 _mm512_storeu_ps(y + 0, vf0);
83 y += 16;
84 }
85
86 for (; elements >= 16 * sizeof(float); elements -= 16 * sizeof(float)) {
87 // Load 16 inputs at a time.
88 const __m512 vx = _mm512_loadu_ps(x);
89 x += 16;
90
91 // Compute reduced argument elements := round(x / log(2)).
92 const __m512 vn = _mm512_roundscale_ps(_mm512_mul_ps(vx, vlog2e), 0);
93
94 // Compute reduced argument t := x - elements * log(2).
95 // Use Cody-Waite range reduction method (note two constants to represent log(2)) to improve accuracy.
96 __m512 vt = _mm512_fmadd_ps(vn, vminus_ln2_hi, vx);
97 vt = _mm512_fmadd_ps(vn, vminus_ln2_lo, vt);
98
99 // Compute degree-5 polynomial approximation for exp(t) on [-log(2)/2, log(2)/2].
100 __m512 vp = _mm512_fmadd_ps(vc5, vt, vc4);
101 vp = _mm512_fmadd_ps(vp, vt, vc3);
102 vp = _mm512_fmadd_ps(vp, vt, vc2);
103 vp = _mm512_fmadd_ps(vp, vt, vc1);
104 vp = _mm512_fmadd_ps(vp, vt, vc0);
105
106 // Multiply "extended" floating-point numbers in ("mantissa", "exponent") representation.
107 __m512 vf = _mm512_mul_ps(vp, vscalev);
108 const __m512 ve = _mm512_add_ps(vn, vscalee);
109
110 // Multiply "mantissa" by the exp2("exponent").
111 vf = _mm512_scalef_ps(vf, ve);
112
113 // Store 16 results at a time.
114 _mm512_storeu_ps(y, vf);
115 y += 16;
116 }
117 if XNN_UNLIKELY(elements != 0) {
118 // Prepare mask for valid 32-bit elements (depends on elements).
119 elements >>= 2 /* log2(sizeof(float)) */;
120 const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << elements) - UINT32_C(1)));
121
122 // Load up to 15 inputs at a time.
123 const __m512 vx = _mm512_maskz_loadu_ps(vmask, x);
124
125 // Compute reduced argument elements := round(x / log(2)).
126 const __m512 vn = _mm512_roundscale_ps(_mm512_mul_ps(vx, vlog2e), 0);
127
128 // Compute reduced argument t := x - elements * log(2).
129 // Use Cody-Waite range reduction method (note two constants to represent log(2)) to improve accuracy.
130 __m512 vt = _mm512_fmadd_ps(vn, vminus_ln2_hi, vx);
131 vt = _mm512_fmadd_ps(vn, vminus_ln2_lo, vt);
132
133 // Compute degree-5 polynomial approximation for exp(t) on [-log(2)/2, log(2)/2].
134 __m512 vp = _mm512_fmadd_ps(vc5, vt, vc4);
135 vp = _mm512_fmadd_ps(vp, vt, vc3);
136 vp = _mm512_fmadd_ps(vp, vt, vc2);
137 vp = _mm512_fmadd_ps(vp, vt, vc1);
138 vp = _mm512_fmadd_ps(vp, vt, vc0);
139
140 // Multiply "extended" floating-point numbers in ("mantissa", "exponent") representation.
141 __m512 vf = _mm512_mul_ps(vp, vscalev);
142 const __m512 ve = _mm512_add_ps(vn, vscalee);
143
144 // Multiply "mantissa" by the exp2("exponent").
145 vf = _mm512_scalef_ps(vf, ve);
146
147 // Store up to 15 results at a time.
148 _mm512_mask_storeu_ps(y, vmask, vf);
149 }
150 _mm256_zeroupper();
151 }
152