|
|
|
1 |
|
/** |
2 |
|
*** Copyright (C) 1985-2007 Intel Corporation. All rights reserved. |
3 |
|
*** |
4 |
|
*** The information and source code contained herein is the exclusive |
5 |
|
*** property of Intel Corporation and may not be disclosed, examined |
6 |
|
*** or reproduced in whole or in part without explicit written authorization |
7 |
|
*** from the company. |
8 |
|
*** |
9 |
|
**/ |
10 |
|
|
11 |
|
/* |
12 |
|
* smmintrin.h |
13 |
|
* |
14 |
|
* Principal header file for Intel(R) Core(TM) 2 Duo processor |
15 |
|
* SSE4.1 intrinsics |
16 |
|
*/ |
17 |
|
|
18 |
|
#pragma once |
19 |
|
#ifndef __midl |
20 |
|
#ifndef _INCLUDED_SMM |
21 |
|
#define _INCLUDED_SMM |
22 |
|
|
23 |
|
#if defined(_M_CEE_PURE) |
24 |
|
#error ERROR: EMM intrinsics not supported in the pure mode! |
25 |
|
#else |
26 |
|
|
27 |
|
#include <tmmintrin.h> |
28 |
|
|
29 |
|
|
30 |
|
/* |
31 |
|
* Rounding mode macros |
32 |
|
*/ |
33 |
|
|
34 |
|
#define _MM_FROUND_TO_NEAREST_INT 0x00 |
35 |
|
#define _MM_FROUND_TO_NEG_INF 0x01 |
36 |
|
#define _MM_FROUND_TO_POS_INF 0x02 |
37 |
|
#define _MM_FROUND_TO_ZERO 0x03 |
38 |
|
#define _MM_FROUND_CUR_DIRECTION 0x04 |
39 |
|
|
40 |
|
#define _MM_FROUND_RAISE_EXC 0x00 |
41 |
|
#define _MM_FROUND_NO_EXC 0x08 |
42 |
|
|
43 |
|
#define _MM_FROUND_NINT _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_RAISE_EXC |
44 |
|
#define _MM_FROUND_FLOOR _MM_FROUND_TO_NEG_INF | _MM_FROUND_RAISE_EXC |
45 |
|
#define _MM_FROUND_CEIL _MM_FROUND_TO_POS_INF | _MM_FROUND_RAISE_EXC |
46 |
|
#define _MM_FROUND_TRUNC _MM_FROUND_TO_ZERO | _MM_FROUND_RAISE_EXC |
47 |
|
#define _MM_FROUND_RINT _MM_FROUND_CUR_DIRECTION | _MM_FROUND_RAISE_EXC |
48 |
|
#define _MM_FROUND_NEARBYINT _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC |
49 |
|
|
50 |
|
/* |
51 |
|
* MACRO functions for ceil/floor intrinsics |
52 |
|
*/ |
53 |
|
|
54 |
|
#define _mm_ceil_pd(val) _mm_round_pd((val), _MM_FROUND_CEIL); |
55 |
|
#define _mm_ceil_sd(dst, val) _mm_round_sd((dst), (val), _MM_FROUND_CEIL); |
56 |
|
|
57 |
|
#define _mm_floor_pd(val) _mm_round_pd((val), _MM_FROUND_FLOOR); |
58 |
|
#define _mm_floor_sd(dst, val) _mm_round_sd((dst), (val), _MM_FROUND_FLOOR); |
59 |
|
|
60 |
|
#define _mm_ceil_ps(val) _mm_round_ps((val), _MM_FROUND_CEIL); |
61 |
|
#define _mm_ceil_ss(dst, val) _mm_round_ss((dst), (val), _MM_FROUND_CEIL); |
62 |
|
|
63 |
|
#define _mm_floor_ps(val) _mm_round_ps((val), _MM_FROUND_FLOOR); |
64 |
|
#define _mm_floor_ss(dst, val) _mm_round_ss((dst), (val), _MM_FROUND_FLOOR); |
65 |
|
|
66 |
|
#define _mm_test_all_zeros(mask, val) _mm_testz_si128((mask), (val)) |
67 |
|
|
68 |
|
/* |
69 |
|
* MACRO functions for packed integer 128-bit comparison intrinsics. |
70 |
|
*/ |
71 |
|
|
72 |
|
#define _mm_test_all_ones(val) \ |
73 |
|
_mm_testc_si128((val), _mm_cmpeq_epi32((val),(val))) |
74 |
|
|
75 |
|
#define _mm_test_mix_ones_zeros(mask, val) _mm_testnzc_si128((mask), (val)) |
76 |
|
|
77 |
|
#if __cplusplus |
78 |
|
extern "C" { |
79 |
|
#endif |
80 |
|
|
81 |
|
// Integer blend instructions - select data from 2 sources |
82 |
|
// using constant/variable mask |
83 |
|
|
84 |
|
extern __m128i _mm_blend_epi16 (__m128i v1, __m128i v2, |
85 |
|
const int mask); |
86 |
|
extern __m128i _mm_blendv_epi8 (__m128i v1, __m128i v2, __m128i mask); |
87 |
|
|
88 |
|
// Float single precision blend instructions - select data |
89 |
|
// from 2 sources using constant/variable mask |
90 |
|
|
91 |
|
extern __m128 _mm_blend_ps (__m128 v1, __m128 v2, const int mask); |
92 |
|
extern __m128 _mm_blendv_ps(__m128 v1, __m128 v2, __m128 v3); |
93 |
|
|
94 |
|
// Float double precision blend instructions - select data |
95 |
|
// from 2 sources using constant/variable mask |
96 |
|
|
97 |
|
extern __m128d _mm_blend_pd (__m128d v1, __m128d v2, const int mask); |
98 |
|
extern __m128d _mm_blendv_pd(__m128d v1, __m128d v2, __m128d v3); |
99 |
|
|
100 |
|
// Dot product instructions with mask-defined summing and zeroing |
101 |
|
// of result's parts |
102 |
|
|
103 |
|
extern __m128 _mm_dp_ps(__m128 val1, __m128 val2, const int mask); |
104 |
|
extern __m128d _mm_dp_pd(__m128d val1, __m128d val2, const int mask); |
105 |
|
|
106 |
|
// Packed integer 64-bit comparison, zeroing or filling with ones |
107 |
|
// corresponding parts of result |
108 |
|
|
109 |
|
extern __m128i _mm_cmpeq_epi64(__m128i val1, __m128i val2); |
110 |
|
|
111 |
|
// Min/max packed integer instructions |
112 |
|
|
113 |
|
extern __m128i _mm_min_epi8 (__m128i val1, __m128i val2); |
114 |
|
extern __m128i _mm_max_epi8 (__m128i val1, __m128i val2); |
115 |
|
|
116 |
|
extern __m128i _mm_min_epu16(__m128i val1, __m128i val2); |
117 |
|
extern __m128i _mm_max_epu16(__m128i val1, __m128i val2); |
118 |
|
|
119 |
|
extern __m128i _mm_min_epi32(__m128i val1, __m128i val2); |
120 |
|
extern __m128i _mm_max_epi32(__m128i val1, __m128i val2); |
121 |
|
extern __m128i _mm_min_epu32(__m128i val1, __m128i val2); |
122 |
|
extern __m128i _mm_max_epu32(__m128i val1, __m128i val2); |
123 |
|
|
124 |
|
// Packed integer 32-bit multiplication with truncation |
125 |
|
// of upper halves of results |
126 |
|
|
137 |
|
extern int _mm_testz_si128(__m128i mask, __m128i val); |
138 |
|
|
139 |
|
// Packed integer 128-bit bitwise comparison. |
140 |
|
// return 1 if (val 'and_not' mask) == 0 |
141 |
|
|
142 |
|
extern int _mm_testc_si128(__m128i mask, __m128i val); |
143 |
|
|
144 |
|
// Packed integer 128-bit bitwise comparison |
145 |
|
// ZF = ((val 'and' mask) == 0) CF = ((val 'and_not' mask) == 0) |
146 |
|
// return 1 if both ZF and CF are 0 |
147 |
|
|
148 |
|
extern int _mm_testnzc_si128(__m128i mask, __m128i s2); |
149 |
|
|
150 |
|
// Insert single precision float into packed single precision |
151 |
|
// array element selected by index. |
152 |
|
// The bits [7-6] of the 3d parameter define src index, |
153 |
|
// the bits [5-4] define dst index, and bits [3-0] define zeroing |
154 |
|
// mask for dst |
155 |
|
|
156 |
|
extern __m128 _mm_insert_ps(__m128 dst, __m128 src, const int ndx); |
157 |
|
|
158 |
|
// Helper macro to create ndx-parameter value for _mm_insert_ps |
159 |
|
|
160 |
|
#define _MM_MK_INSERTPS_NDX(srcField, dstField, zeroMask) \ |
161 |
|
(((srcField)<<6) | ((dstField)<<4) | (zeroMask)) |
162 |
|
|
163 |
|
// Extract binary representation of single precision float from |
164 |
|
// packed single precision array element selected by index |
165 |
|
|
166 |
|
extern int _mm_extract_ps(__m128 src, const int ndx); |
167 |
|
|
168 |
|
// Extract single precision float from packed single precision |
169 |
|
// array element selected by index into dest |
170 |
|
|
171 |
|
#define _MM_EXTRACT_FLOAT(dest, src, ndx) \ |
172 |
|
*((int*)&(dest)) = _mm_extract_ps((src), (ndx)) |
173 |
|
|
174 |
|
// Extract specified single precision float element |
175 |
|
// into the lower part of __m128 |
176 |
|
|
177 |
|
#define _MM_PICK_OUT_PS(src, num) \ |
178 |
|
_mm_insert_ps(_mm_setzero_ps(), (src), \ |
179 |
|
_MM_MK_INSERTPS_NDX((num), 0, 0x0e)); |
180 |
|
|
181 |
|
// Insert integer into packed integer array element |
182 |
|
// selected by index |
183 |
|
|
184 |
|
extern __m128i _mm_insert_epi8 (__m128i dst, int s, const int ndx); |
185 |
|
extern __m128i _mm_insert_epi32(__m128i dst, int s, const int ndx); |
186 |
|
|
187 |
|
#if defined(_M_X64) |
188 |
|
extern __m128i _mm_insert_epi64(__m128i dst, __int64 s, const int ndx); |
189 |
|
#endif |
190 |
|
// Extract integer from packed integer array element |
191 |
|
// selected by index |
192 |
|
|
193 |
|
extern int _mm_extract_epi8 (__m128i src, const int ndx); |
194 |
|
extern int _mm_extract_epi32(__m128i src, const int ndx); |
195 |
|
|
196 |
|
#if defined(_M_X64) |
197 |
|
extern __int64 _mm_extract_epi64(__m128i src, const int ndx); |
198 |
|
#endif |
199 |
|
|
200 |
|
// Horizontal packed word minimum and its index in |
201 |
|
// result[15:0] and result[18:16] respectively |
202 |
|
|
203 |
|
extern __m128i _mm_minpos_epu16(__m128i shortValues); |
204 |
|
|
205 |
|
// Packed/single float double precision rounding |
206 |
|
|
207 |
|
extern __m128d _mm_round_pd(__m128d val, int iRoundMode); |
208 |
|
extern __m128d _mm_round_sd(__m128d dst, __m128d val, int iRoundMode); |
209 |
|
|
210 |
|
// Packed/single float single precision rounding |
211 |
|
|
212 |
|
extern __m128 _mm_round_ps(__m128 val, int iRoundMode); |
213 |
|
extern __m128 _mm_round_ss(__m128 dst, __m128 val, int iRoundMode); |
214 |
|
|
215 |
|
// Packed integer sign-extension |
216 |
|
|
217 |
|
extern __m128i _mm_cvtepi8_epi32 (__m128i byteValues); |
218 |
|
extern __m128i _mm_cvtepi16_epi32(__m128i shortValues); |
219 |
|
extern __m128i _mm_cvtepi8_epi64 (__m128i byteValues); |
220 |
|
extern __m128i _mm_cvtepi32_epi64(__m128i intValues); |
221 |
|
extern __m128i _mm_cvtepi16_epi64(__m128i shortValues); |
222 |
|
extern __m128i _mm_cvtepi8_epi16 (__m128i byteValues); |
223 |
|
|
224 |
|
// Packed integer zero-extension |
225 |
|
|
226 |
|
extern __m128i _mm_cvtepu8_epi32 (__m128i byteValues); |
227 |
|
extern __m128i _mm_cvtepu16_epi32(__m128i shortValues); |
228 |
|
extern __m128i _mm_cvtepu8_epi64 (__m128i shortValues); |
229 |
|
extern __m128i _mm_cvtepu32_epi64(__m128i intValues); |
230 |
|
extern __m128i _mm_cvtepu16_epi64(__m128i shortValues); |
231 |
|
extern __m128i _mm_cvtepu8_epi16 (__m128i byteValues); |
232 |
|
|
233 |
|
|
234 |
|
// Pack 8 double words from 2 operands into 8 words of result |
235 |
|
// with unsigned saturation |
236 |
|
|
237 |
|
extern __m128i _mm_packus_epi32(__m128i val1, __m128i val2); |
238 |
|
|
239 |
|
// Sum absolute 8-bit integer difference of adjacent groups of 4 byte |
240 |
|
// integers in operands. Starting offsets within operands are |
241 |
|
// determined by mask |
242 |
|
|
243 |
|
extern __m128i _mm_mpsadbw_epu8(__m128i s1, __m128i s2, const int msk); |
244 |
|
|
245 |
|
/* |
246 |
|
* Load double quadword using non-temporal aligned hint |
247 |
|
*/ |
248 |
|
|
249 |
|
extern __m128i _mm_stream_load_si128(__m128i* v1); |
250 |
|
|
251 |
|
#if defined __cplusplus |
252 |
|
}; /* End "C" */ |
253 |
|
#endif /* __cplusplus */ |
254 |
|
|
255 |
|
#endif /* defined(_M_CEE_PURE) */ |
256 |
|
|
257 |
|
#endif |
258 |
|
#endif /* _INCLUDED_SMM */ |
259 |
|
|
|
|
|