1/**
2 * This file has no copyright assigned and is placed in the Public Domain.
3 * This file is part of the mingw-w64 runtime package.
4 * No warranty is given; refer to the file DISCLAIMER.PD within this package.
5 */
6#ifndef _FVEC_H_INCLUDED
7#define _FVEC_H_INCLUDED
8
9#ifndef RC_INVOKED
10#ifndef __cplusplus
11#error ERROR: This file is only supported in C++ compilations!
12#endif
13
14#include <intrin.h>
15#include <assert.h>
16#include <ivec.h>
17#include <crtdefs.h>
18
19#if defined(_ENABLE_VEC_DEBUG)
20#include <iostream>
21#endif
22
23#pragma pack(push,_CRT_PACKING)
24
25#ifdef __SSE__
26
27#pragma pack(push,16)
28
29#define EXPLICIT explicit
30
31class F32vec4 {
32protected:
33  __m128 vec;
34public:
35  F32vec4() {}
36  F32vec4(__m128 m) { vec = m;}
37  F32vec4(float f3,float f2,float f1,float f0) { vec= _mm_set_ps(f3,f2,f1,f0); }
38  EXPLICIT F32vec4(float f) { vec = _mm_set_ps1(f); }
39  EXPLICIT F32vec4(double d) { vec = _mm_set_ps1((float) d); }
40  F32vec4& operator =(float f) { vec = _mm_set_ps1(f); return *this; }
41  F32vec4& operator =(double d) { vec = _mm_set_ps1((float) d); return *this; }
42  operator __m128() const { return vec; }
43  friend F32vec4 operator &(const F32vec4 &a,const F32vec4 &b) { return _mm_and_ps(a,b); }
44  friend F32vec4 operator |(const F32vec4 &a,const F32vec4 &b) { return _mm_or_ps(a,b); }
45  friend F32vec4 operator ^(const F32vec4 &a,const F32vec4 &b) { return _mm_xor_ps(a,b); }
46  friend F32vec4 operator +(const F32vec4 &a,const F32vec4 &b) { return _mm_add_ps(a,b); }
47  friend F32vec4 operator -(const F32vec4 &a,const F32vec4 &b) { return _mm_sub_ps(a,b); }
48  friend F32vec4 operator *(const F32vec4 &a,const F32vec4 &b) { return _mm_mul_ps(a,b); }
49  friend F32vec4 operator /(const F32vec4 &a,const F32vec4 &b) { return _mm_div_ps(a,b); }
50  F32vec4& operator =(const F32vec4 &a) { vec = a.vec; return *this; }
51  F32vec4& operator =(const __m128 &avec) { vec = avec; return *this; }
52  F32vec4& operator +=(F32vec4 &a) { return *this = _mm_add_ps(vec,a); }
53  F32vec4& operator -=(F32vec4 &a) { return *this = _mm_sub_ps(vec,a); }
54  F32vec4& operator *=(F32vec4 &a) { return *this = _mm_mul_ps(vec,a); }
55  F32vec4& operator /=(F32vec4 &a) { return *this = _mm_div_ps(vec,a); }
56  F32vec4& operator &=(F32vec4 &a) { return *this = _mm_and_ps(vec,a); }
57  F32vec4& operator |=(F32vec4 &a) { return *this = _mm_or_ps(vec,a); }
58  F32vec4& operator ^=(F32vec4 &a) { return *this = _mm_xor_ps(vec,a); }
59  friend float add_horizontal(F32vec4 &a) {
60    F32vec4 ftemp = _mm_add_ss(a,_mm_add_ss(_mm_shuffle_ps(a,a,1),_mm_add_ss(_mm_shuffle_ps(a,a,2),_mm_shuffle_ps(a,a,3))));
61    return ftemp[0];
62  }
63  friend F32vec4 sqrt(const F32vec4 &a) { return _mm_sqrt_ps(a); }
64  friend F32vec4 rcp(const F32vec4 &a) { return _mm_rcp_ps(a); }
65  friend F32vec4 rsqrt(const F32vec4 &a) { return _mm_rsqrt_ps(a); }
66  friend F32vec4 rcp_nr(const F32vec4 &a) {
67    F32vec4 Ra0 = _mm_rcp_ps(a);
68    return _mm_sub_ps(_mm_add_ps(Ra0,Ra0),_mm_mul_ps(_mm_mul_ps(Ra0,a),Ra0));
69  }
70  friend F32vec4 rsqrt_nr(const F32vec4 &a) {
71    static const F32vec4 fvecf0pt5(0.5f);
72    static const F32vec4 fvecf3pt0(3.0f);
73    F32vec4 Ra0 = _mm_rsqrt_ps(a);
74    return (fvecf0pt5 *Ra0) *(fvecf3pt0 - (a *Ra0) *Ra0);
75
76  }
77#define Fvec32s4_COMP(op) friend F32vec4 cmp##op (const F32vec4 &a,const F32vec4 &b) { return _mm_cmp##op##_ps(a,b); }
78  Fvec32s4_COMP(eq)
79    Fvec32s4_COMP(lt)
80    Fvec32s4_COMP(le)
81    Fvec32s4_COMP(gt)
82    Fvec32s4_COMP(ge)
83    Fvec32s4_COMP(neq)
84    Fvec32s4_COMP(nlt)
85    Fvec32s4_COMP(nle)
86    Fvec32s4_COMP(ngt)
87    Fvec32s4_COMP(nge)
88#undef Fvec32s4_COMP
89
90    friend F32vec4 simd_min(const F32vec4 &a,const F32vec4 &b) { return _mm_min_ps(a,b); }
91  friend F32vec4 simd_max(const F32vec4 &a,const F32vec4 &b) { return _mm_max_ps(a,b); }
92
93#if defined(_ENABLE_VEC_DEBUG)
94  friend std::ostream & operator<<(std::ostream & os,const F32vec4 &a) {
95    float *fp = (float*)&a;
96    os << "[3]:" << *(fp+3)
97      << " [2]:" << *(fp+2)
98      << " [1]:" << *(fp+1)
99      << " [0]:" << *fp;
100    return os;
101  }
102#endif
103  const float& operator[](int i) const {
104    assert((0 <= i) && (i <= 3));
105    float *fp = (float*)&vec;
106    return *(fp+i);
107  }
108  float& operator[](int i) {
109    assert((0 <= i) && (i <= 3));
110    float *fp = (float*)&vec;
111    return *(fp+i);
112  }
113};
114
115inline F32vec4 unpack_low(const F32vec4 &a,const F32vec4 &b) { return _mm_unpacklo_ps(a,b); }
116inline F32vec4 unpack_high(const F32vec4 &a,const F32vec4 &b) { return _mm_unpackhi_ps(a,b); }
117inline int move_mask(const F32vec4 &a) { return _mm_movemask_ps(a); }
118inline void loadu(F32vec4 &a,float *p) { a = _mm_loadu_ps(p); }
119inline void storeu(float *p,const F32vec4 &a) { _mm_storeu_ps(p,a); }
120inline void store_nta(float *p,F32vec4 &a) { _mm_stream_ps(p,a); }
121
122#define Fvec32s4_SELECT(op) inline F32vec4 select_##op (const F32vec4 &a,const F32vec4 &b,const F32vec4 &c,const F32vec4 &d) { F32vec4 mask = _mm_cmp##op##_ps(a,b); return((mask & c) | F32vec4((_mm_andnot_ps(mask,d)))); }
123Fvec32s4_SELECT(eq)
124Fvec32s4_SELECT(lt)
125Fvec32s4_SELECT(le)
126Fvec32s4_SELECT(gt)
127Fvec32s4_SELECT(ge)
128Fvec32s4_SELECT(neq)
129Fvec32s4_SELECT(nlt)
130Fvec32s4_SELECT(nle)
131Fvec32s4_SELECT(ngt)
132Fvec32s4_SELECT(nge)
133#undef Fvec32s4_SELECT
134
135#if 0 /* Commented until required types are defined */
136inline Is16vec4 simd_max(const Is16vec4 &a,const Is16vec4 &b) { return _m_pmaxsw(a,b); }
137inline Is16vec4 simd_min(const Is16vec4 &a,const Is16vec4 &b) { return _m_pminsw(a,b); }
138inline Iu8vec8 simd_max(const Iu8vec8 &a,const Iu8vec8 &b) { return _m_pmaxub(a,b); }
139inline Iu8vec8 simd_min(const Iu8vec8 &a,const Iu8vec8 &b) { return _m_pminub(a,b); }
140inline Iu16vec4 simd_avg(const Iu16vec4 &a,const Iu16vec4 &b) { return _m_pavgw(a,b); }
141inline Iu8vec8 simd_avg(const Iu8vec8 &a,const Iu8vec8 &b) { return _m_pavgb(a,b); }
142inline int move_mask(const I8vec8 &a) { return _m_pmovmskb(a); }
143inline Iu16vec4 mul_high(const Iu16vec4 &a,const Iu16vec4 &b) { return _m_pmulhuw(a,b); }
144inline void mask_move(const I8vec8 &a,const I8vec8 &b,char *addr) { _m_maskmovq(a,b,addr); }
145inline void store_nta(__m64 *p,M64 &a) { _mm_stream_pi(p,a); }
146inline int F32vec4ToInt(const F32vec4 &a) { return _mm_cvtt_ss2si(a); }
147inline Is32vec2 F32vec4ToIs32vec2 (const F32vec4 &a) {
148  __m64 result;
149  result = _mm_cvtt_ps2pi(a);
150  return Is32vec2(result);
151}
152#endif
153
154inline F32vec4 IntToF32vec4(const F32vec4 &a,int i) {
155  __m128 result;
156  result = _mm_cvt_si2ss(a,i);
157  return F32vec4(result);
158}
159
160#if 0 /* Commented until required types are defined */
161inline F32vec4 Is32vec2ToF32vec4(const F32vec4 &a,const Is32vec2 &b) {
162  __m128 result;
163  result = _mm_cvt_pi2ps(a,b);
164  return F32vec4(result);
165}
166#endif
167
168class F32vec1 {
169protected:
170  __m128 vec;
171public:
172  F32vec1() {}
173  F32vec1(int i) { vec = _mm_cvt_si2ss(vec,i);};
174  EXPLICIT F32vec1(float f) { vec = _mm_set_ss(f); }
175  EXPLICIT F32vec1(double d) { vec = _mm_set_ss((float) d); }
176  F32vec1(__m128 m) { vec = m; }
177  operator __m128() const { return vec; }
178  friend F32vec1 operator &(const F32vec1 &a,const F32vec1 &b) { return _mm_and_ps(a,b); }
179  friend F32vec1 operator |(const F32vec1 &a,const F32vec1 &b) { return _mm_or_ps(a,b); }
180  friend F32vec1 operator ^(const F32vec1 &a,const F32vec1 &b) { return _mm_xor_ps(a,b); }
181  friend F32vec1 operator +(const F32vec1 &a,const F32vec1 &b) { return _mm_add_ss(a,b); }
182  friend F32vec1 operator -(const F32vec1 &a,const F32vec1 &b) { return _mm_sub_ss(a,b); }
183  friend F32vec1 operator *(const F32vec1 &a,const F32vec1 &b) { return _mm_mul_ss(a,b); }
184  friend F32vec1 operator /(const F32vec1 &a,const F32vec1 &b) { return _mm_div_ss(a,b); }
185  F32vec1& operator +=(F32vec1 &a) { return *this = _mm_add_ss(vec,a); }
186  F32vec1& operator -=(F32vec1 &a) { return *this = _mm_sub_ss(vec,a); }
187  F32vec1& operator *=(F32vec1 &a) { return *this = _mm_mul_ss(vec,a); }
188  F32vec1& operator /=(F32vec1 &a) { return *this = _mm_div_ss(vec,a); }
189  F32vec1& operator &=(F32vec1 &a) { return *this = _mm_and_ps(vec,a); }
190  F32vec1& operator |=(F32vec1 &a) { return *this = _mm_or_ps(vec,a); }
191  F32vec1& operator ^=(F32vec1 &a) { return *this = _mm_xor_ps(vec,a); }
192  friend F32vec1 sqrt(const F32vec1 &a) { return _mm_sqrt_ss(a); }
193  friend F32vec1 rcp(const F32vec1 &a) { return _mm_rcp_ss(a); }
194  friend F32vec1 rsqrt(const F32vec1 &a) { return _mm_rsqrt_ss(a); }
195  friend F32vec1 rcp_nr(const F32vec1 &a) {
196    F32vec1 Ra0 = _mm_rcp_ss(a);
197    return _mm_sub_ss(_mm_add_ss(Ra0,Ra0),_mm_mul_ss(_mm_mul_ss(Ra0,a),Ra0));
198  }
199  friend F32vec1 rsqrt_nr(const F32vec1 &a) {
200    static const F32vec1 fvecf0pt5(0.5f);
201    static const F32vec1 fvecf3pt0(3.0f);
202    F32vec1 Ra0 = _mm_rsqrt_ss(a);
203    return (fvecf0pt5 *Ra0) *(fvecf3pt0 - (a *Ra0) *Ra0);
204  }
205#define Fvec32s1_COMP(op) friend F32vec1 cmp##op (const F32vec1 &a,const F32vec1 &b) { return _mm_cmp##op##_ss(a,b); }
206  Fvec32s1_COMP(eq)
207    Fvec32s1_COMP(lt)
208    Fvec32s1_COMP(le)
209    Fvec32s1_COMP(gt)
210    Fvec32s1_COMP(ge)
211    Fvec32s1_COMP(neq)
212    Fvec32s1_COMP(nlt)
213    Fvec32s1_COMP(nle)
214    Fvec32s1_COMP(ngt)
215    Fvec32s1_COMP(nge)
216#undef Fvec32s1_COMP
217
218    friend F32vec1 simd_min(const F32vec1 &a,const F32vec1 &b) { return _mm_min_ss(a,b); }
219  friend F32vec1 simd_max(const F32vec1 &a,const F32vec1 &b) { return _mm_max_ss(a,b); }
220
221#if defined(_ENABLE_VEC_DEBUG)
222  friend std::ostream & operator<<(std::ostream & os,const F32vec1 &a) {
223    float *fp = (float*)&a;
224    os << "float:" << *fp;
225    return os;
226  }
227#endif
228};
229
230#define Fvec32s1_SELECT(op) inline F32vec1 select_##op (const F32vec1 &a,const F32vec1 &b,const F32vec1 &c,const F32vec1 &d) { F32vec1 mask = _mm_cmp##op##_ss(a,b); return((mask & c) | F32vec1((_mm_andnot_ps(mask,d)))); }
231Fvec32s1_SELECT(eq)
232Fvec32s1_SELECT(lt)
233Fvec32s1_SELECT(le)
234Fvec32s1_SELECT(gt)
235Fvec32s1_SELECT(ge)
236Fvec32s1_SELECT(neq)
237Fvec32s1_SELECT(nlt)
238Fvec32s1_SELECT(nle)
239Fvec32s1_SELECT(ngt)
240Fvec32s1_SELECT(nge)
241#undef Fvec32s1_SELECT
242
243inline int F32vec1ToInt(const F32vec1 &a)
244{
245  return _mm_cvtt_ss2si(a);
246}
247
248#pragma pack(pop)
249
250#endif /* #ifdef __SSE__ */
251#pragma pack(pop)
252#endif
253#endif
254