smmintrin.h revision 208954 
1/*===---- smmintrin.h - SSE4 intrinsics ------------------------------------===
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to deal
5 * in the Software without restriction, including without limitation the rights
6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7 * copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19 * THE SOFTWARE.
20 *
21 *===-----------------------------------------------------------------------===
22 */
23
24#ifndef _SMMINTRIN_H
25#define _SMMINTRIN_H
26
27#ifndef __SSE4_1__
28#error "SSE4.1 instruction set not enabled"
29#else
30
31#include <tmmintrin.h>
32
33/* Type defines.  */
34typedef double __v2df __attribute__ ((__vector_size__ (16)));
35typedef long long __v2di __attribute__ ((__vector_size__ (16)));
36
37/* SSE4 Rounding macros. */
38#define _MM_FROUND_TO_NEAREST_INT    0x00
39#define _MM_FROUND_TO_NEG_INF        0x01
40#define _MM_FROUND_TO_POS_INF        0x02
41#define _MM_FROUND_TO_ZERO           0x03
42#define _MM_FROUND_CUR_DIRECTION     0x04
43
44#define _MM_FROUND_RAISE_EXC         0x00
45#define _MM_FROUND_NO_EXC            0x08
46
47#define _MM_FROUND_NINT      (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEAREST_INT)
48#define _MM_FROUND_FLOOR     (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEG_INF)
49#define _MM_FROUND_CEIL      (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_POS_INF)
50#define _MM_FROUND_TRUNC     (_MM_FROUND_RAISE_EXC | _MM_FROUND_TO_ZERO)
51#define _MM_FROUND_RINT      (_MM_FROUND_RAISE_EXC | _MM_FROUND_CUR_DIRECTION)
52#define _MM_FROUND_NEARBYINT (_MM_FROUND_NO_EXC | _MM_FROUND_CUR_DIRECTION)
53
54#define _mm_ceil_ps(X)       _mm_round_ps((X), _MM_FROUND_CEIL)
55#define _mm_ceil_pd(X)       _mm_round_pd((X), _MM_FROUND_CEIL)
56#define _mm_ceil_ss(X, Y)    _mm_round_ss((X), (Y), _MM_FROUND_CEIL)
57#define _mm_ceil_sd(X, Y)    _mm_round_sd((X), (Y), _MM_FROUND_CEIL)
58
59#define _mm_floor_ps(X)      _mm_round_ps((X), _MM_FROUND_FLOOR)
60#define _mm_floor_pd(X)      _mm_round_pd((X), _MM_FROUND_FLOOR)
61#define _mm_floor_ss(X, Y)   _mm_round_ss((X), (Y), _MM_FROUND_FLOOR)
62#define _mm_floor_sd(X, Y)   _mm_round_sd((X), (Y), _MM_FROUND_FLOOR)
63
64#define _mm_round_ps(X, Y)      __builtin_ia32_roundps((X), (Y))
65#define _mm_round_ss(X, Y, M)   __builtin_ia32_roundss((X), (Y), (M))
66#define _mm_round_pd(X, M)      __builtin_ia32_roundpd((X), (M))
67#define _mm_round_sd(X, Y, M)   __builtin_ia32_roundsd((X), (Y), (M))
68
69/* SSE4 Packed Blending Intrinsics.  */
70static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
71_mm_blend_pd (__m128d __V1, __m128d __V2, const int __M)
72{
73  return (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, __M);
74}
75
76static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
77_mm_blend_ps (__m128 __V1, __m128 __V2, const int __M)
78{
79  return (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, __M);
80}
81
82static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
83_mm_blendv_pd (__m128d __V1, __m128d __V2, __m128d __M)
84{
85  return (__m128d) __builtin_ia32_blendvpd ((__v2df)__V1, (__v2df)__V2,
86                                            (__v2df)__M);
87}
88
89static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
90_mm_blendv_ps (__m128 __V1, __m128 __V2, __m128 __M)
91{
92  return (__m128) __builtin_ia32_blendvps ((__v4sf)__V1, (__v4sf)__V2,
93                                           (__v4sf)__M);
94}
95
96static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
97_mm_blendv_epi8 (__m128i __V1, __m128i __V2, __m128i __M)
98{
99  return (__m128i) __builtin_ia32_pblendvb128 ((__v16qi)__V1, (__v16qi)__V2,
100                                               (__v16qi)__M);
101}
102
103static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
104_mm_blend_epi16 (__m128i __V1, __m128i __V2, const int __M)
105{
106  return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, __M);
107}
108
109/* SSE4 Dword Multiply Instructions.  */
110static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
111_mm_mullo_epi32 (__m128i __V1, __m128i __V2)
112{
113  return (__m128i) ((__v4si)__V1 * (__v4si)__V2);
114}
115
116static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
117_mm_mul_epi32 (__m128i __V1, __m128i __V2)
118{
119  return (__m128i) __builtin_ia32_pmuldq128 ((__v4si)__V1, (__v4si)__V2);
120}
121
122/* SSE4 Floating Point Dot Product Instructions.  */
123#define _mm_dp_ps(X, Y, M) __builtin_ia32_dpps ((X), (Y), (M))
124#define _mm_dp_pd(X, Y, M) __builtin_ia32_dppd ((X), (Y), (M))
125
126/* SSE4 Streaming Load Hint Instruction.  */
127static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
128_mm_stream_load_si128 (__m128i *__V)
129{
130  return (__m128i) __builtin_ia32_movntdqa ((__v2di *) __V);
131}
132
133/* SSE4 Packed Integer Min/Max Instructions.  */
134static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
135_mm_min_epi8 (__m128i __V1, __m128i __V2)
136{
137  return (__m128i) __builtin_ia32_pminsb128 ((__v16qi) __V1, (__v16qi) __V2);
138}
139
140static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
141_mm_max_epi8 (__m128i __V1, __m128i __V2)
142{
143  return (__m128i) __builtin_ia32_pmaxsb128 ((__v16qi) __V1, (__v16qi) __V2);
144}
145
146static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
147_mm_min_epu16 (__m128i __V1, __m128i __V2)
148{
149  return (__m128i) __builtin_ia32_pminuw128 ((__v8hi) __V1, (__v8hi) __V2);
150}
151
152static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
153_mm_max_epu16 (__m128i __V1, __m128i __V2)
154{
155  return (__m128i) __builtin_ia32_pmaxuw128 ((__v8hi) __V1, (__v8hi) __V2);
156}
157
158static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
159_mm_min_epi32 (__m128i __V1, __m128i __V2)
160{
161  return (__m128i) __builtin_ia32_pminsd128 ((__v4si) __V1, (__v4si) __V2);
162}
163
164static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
165_mm_max_epi32 (__m128i __V1, __m128i __V2)
166{
167  return (__m128i) __builtin_ia32_pmaxsd128 ((__v4si) __V1, (__v4si) __V2);
168}
169
170static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
171_mm_min_epu32 (__m128i __V1, __m128i __V2)
172{
173  return (__m128i) __builtin_ia32_pminud128((__v4si) __V1, (__v4si) __V2);
174}
175
176static __inline__  __m128i __attribute__((__always_inline__, __nodebug__))
177_mm_max_epu32 (__m128i __V1, __m128i __V2)
178{
179  return (__m128i) __builtin_ia32_pmaxud128((__v4si) __V1, (__v4si) __V2);
180}
181
182/* SSE4 Insertion and Extraction from XMM Register Instructions.  */
183#define _mm_insert_ps(X, Y, N) __builtin_ia32_insertps128((X), (Y), (N))
184#define _mm_extract_ps(X, N) (__extension__                      \
185                              ({ union { int i; float f; } __t;  \
186                                 __v4sf __a = (__v4sf)X;         \
187                                 __t.f = __a[N];                 \
188                                 __t.i;}))
189
190/* Miscellaneous insert and extract macros.  */
191/* Extract a single-precision float from X at index N into D.  */
192#define _MM_EXTRACT_FLOAT(D, X, N) (__extension__ ({ __v4sf __a = (__v4sf)X; \
193                                                    (D) = __a[N]; }))
194
195/* Or together 2 sets of indexes (X and Y) with the zeroing bits (Z) to create
196   an index suitable for _mm_insert_ps.  */
197#define _MM_MK_INSERTPS_NDX(X, Y, Z) (((X) << 6) | ((Y) << 4) | (Z))
198
199/* Extract a float from X at index N into the first index of the return.  */
200#define _MM_PICK_OUT_PS(X, N) _mm_insert_ps (_mm_setzero_ps(), (X),   \
201                                             _MM_MK_INSERTPS_NDX((N), 0, 0x0e))
202
203/* Insert int into packed integer array at index.  */
204#define _mm_insert_epi8(X, I, N) (__extension__ ({ __v16qi __a = (__v16qi)X; \
205                                                   __a[N] = I;               \
206                                                   __a;}))
207#define _mm_insert_epi32(X, I, N) (__extension__ ({ __v4si __a = (__v4si)X; \
208                                                    __a[N] = I;             \
209                                                    __a;}))
210#ifdef __x86_64__
211#define _mm_insert_epi64(X, I, N) (__extension__ ({ __v2di __a = (__v2di)X; \
212                                                    __a[N] = I;             \
213                                                    __a;}))
214#endif /* __x86_64__ */
215
216/* Extract int from packed integer array at index.  */
217#define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)X; \
218                                                 __a[N];}))
219#define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)X; \
220                                                  __a[N];}))
221#ifdef __x86_64__
222#define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)X; \
223                                                  __a[N];}))
224#endif /* __x86_64 */
225
226/* SSE4 128-bit Packed Integer Comparisons.  */
227static __inline__ int __attribute__((__always_inline__, __nodebug__))
228_mm_testz_si128(__m128i __M, __m128i __V)
229{
230  return __builtin_ia32_ptestz128((__v2di)__M, (__v2di)__V);
231}
232
233static __inline__ int __attribute__((__always_inline__, __nodebug__))
234_mm_testc_si128(__m128i __M, __m128i __V)
235{
236  return __builtin_ia32_ptestc128((__v2di)__M, (__v2di)__V);
237}
238
239static __inline__ int __attribute__((__always_inline__, __nodebug__))
240_mm_testnzc_si128(__m128i __M, __m128i __V)
241{
242  return __builtin_ia32_ptestnzc128((__v2di)__M, (__v2di)__V);
243}
244
245#define _mm_test_all_ones(V) _mm_testc_si128((V), _mm_cmpeq_epi32((V), (V)))
246#define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128((M), (V))
247#define _mm_test_all_zeros(M, V) _mm_testz_si128 ((V), (V))
248
249/* SSE4 64-bit Packed Integer Comparisons.  */
250static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
251_mm_cmpeq_epi64(__m128i __V1, __m128i __V2)
252{
253  return (__m128i) __builtin_ia32_pcmpeqq((__v2di)__V1, (__v2di)__V2);
254}
255
256/* SSE4 Packed Integer Sign-Extension.  */
257static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
258_mm_cvtepi8_epi16(__m128i __V)
259{
260  return (__m128i) __builtin_ia32_pmovsxbw128((__v16qi) __V);
261}
262
263static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
264_mm_cvtepi8_epi32(__m128i __V)
265{
266  return (__m128i) __builtin_ia32_pmovsxbd128((__v16qi) __V);
267}
268
269static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
270_mm_cvtepi8_epi64(__m128i __V)
271{
272  return (__m128i) __builtin_ia32_pmovsxbq128((__v16qi) __V);
273}
274
275static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
276_mm_cvtepi16_epi32(__m128i __V)
277{
278  return (__m128i) __builtin_ia32_pmovsxwd128((__v8hi) __V);
279}
280
281static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
282_mm_cvtepi16_epi64(__m128i __V)
283{
284  return (__m128i) __builtin_ia32_pmovsxwq128((__v8hi)__V);
285}
286
287static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
288_mm_cvtepi32_epi64(__m128i __V)
289{
290  return (__m128i) __builtin_ia32_pmovsxdq128((__v4si)__V);
291}
292
293/* SSE4 Packed Integer Zero-Extension.  */
294static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
295_mm_cvtepu8_epi16(__m128i __V)
296{
297  return (__m128i) __builtin_ia32_pmovzxbw128((__v16qi) __V);
298}
299
300static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
301_mm_cvtepu8_epi32(__m128i __V)
302{
303  return (__m128i) __builtin_ia32_pmovzxbd128((__v16qi)__V);
304}
305
306static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
307_mm_cvtepu8_epi64(__m128i __V)
308{
309  return (__m128i) __builtin_ia32_pmovzxbq128((__v16qi)__V);
310}
311
312static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
313_mm_cvtepu16_epi32(__m128i __V)
314{
315  return (__m128i) __builtin_ia32_pmovzxwd128((__v8hi)__V);
316}
317
318static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
319_mm_cvtepu16_epi64(__m128i __V)
320{
321  return (__m128i) __builtin_ia32_pmovzxwq128((__v8hi)__V);
322}
323
324static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
325_mm_cvtepu32_epi64(__m128i __V)
326{
327  return (__m128i) __builtin_ia32_pmovzxdq128((__v4si)__V);
328}
329
330/* SSE4 Pack with Unsigned Saturation.  */
331static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
332_mm_packus_epi32(__m128i __V1, __m128i __V2)
333{
334  return (__m128i) __builtin_ia32_packusdw128((__v4si)__V1, (__v4si)__V2);
335}
336
337/* SSE4 Multiple Packed Sums of Absolute Difference.  */
338#define _mm_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw128((X), (Y), (M))
339
340/* These definitions are normally in nmmintrin.h, but gcc puts them in here
341   so we'll do the same.  */
342#ifdef __SSE4_2__
343
344/* These specify the type of data that we're comparing.  */
345#define _SIDD_UBYTE_OPS                 0x00
346#define _SIDD_UWORD_OPS                 0x01
347#define _SIDD_SBYTE_OPS                 0x02
348#define _SIDD_SWORD_OPS                 0x03
349
350/* These specify the type of comparison operation.  */
351#define _SIDD_CMP_EQUAL_ANY             0x00
352#define _SIDD_CMP_RANGES                0x04
353#define _SIDD_CMP_EQUAL_EACH            0x08
354#define _SIDD_CMP_EQUAL_ORDERED         0x0c
355
356/* These macros specify the polarity of the operation.  */
357#define _SIDD_POSITIVE_POLARITY         0x00
358#define _SIDD_NEGATIVE_POLARITY         0x10
359#define _SIDD_MASKED_POSITIVE_POLARITY  0x20
360#define _SIDD_MASKED_NEGATIVE_POLARITY  0x30
361
362/* These macros are used in _mm_cmpXstri() to specify the return.  */
363#define _SIDD_LEAST_SIGNIFICANT         0x00
364#define _SIDD_MOST_SIGNIFICANT          0x40
365
366/* These macros are used in _mm_cmpXstri() to specify the return.  */
367#define _SIDD_BIT_MASK                  0x00
368#define _SIDD_UNIT_MASK                 0x40
369
370/* SSE4.2 Packed Comparison Intrinsics.  */
371#define _mm_cmpistrm(A, B, M) __builtin_ia32_pcmpistrm128((A), (B), (M))
372#define _mm_cmpistri(A, B, M) __builtin_ia32_pcmpistri128((A), (B), (M))
373
374#define _mm_cmpestrm(A, LA, B, LB, M) \
375     __builtin_ia32_pcmpestrm128((A), (LA), (B), (LB), (M))
376#define _mm_cmpestri(X, LX, Y, LY, M) \
377     __builtin_ia32_pcmpestri128((A), (LA), (B), (LB), (M))
378
379/* SSE4.2 Packed Comparison Intrinsics and EFlag Reading.  */
380#define _mm_cmpistra(A, LA, B, LB, M) \
381     __builtin_ia32_pcmpistria128((A), (LA), (B), (LB), (M))
382#define _mm_cmpistrc(A, LA, B, LB, M) \
383     __builtin_ia32_pcmpistric128((A), (LA), (B), (LB), (M))
384#define _mm_cmpistro(A, LA, B, LB, M) \
385     __builtin_ia32_pcmpistrio128((A), (LA), (B), (LB), (M))
386#define _mm_cmpistrs(A, LA, B, LB, M) \
387     __builtin_ia32_pcmpistris128((A), (LA), (B), (LB), (M))
388#define _mm_cmpistrz(A, LA, B, LB, M) \
389     __builtin_ia32_pcmpistriz128((A), (LA), (B), (LB), (M))
390
391#define _mm_cmpestra(A, LA, B, LB, M) \
392     __builtin_ia32_pcmpestria128((A), (LA), (B), (LB), (M))
393#define _mm_cmpestrc(A, LA, B, LB, M) \
394     __builtin_ia32_pcmpestric128((A), (LA), (B), (LB), (M))
395#define _mm_cmpestro(A, LA, B, LB, M) \
396     __builtin_ia32_pcmpestrio128((A), (LA), (B), (LB), (M))
397#define _mm_cmpestrs(A, LA, B, LB, M) \
398     __builtin_ia32_pcmpestris128((A), (LA), (B), (LB), (M))
399#define _mm_cmpestrz(A, LA, B, LB, M) \
400     __builtin_ia32_pcmpestriz128((A), (LA), (B), (LB), (M))
401
402/* SSE4.2 Compare Packed Data -- Greater Than.  */
403static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
404_mm_cmpgt_epi64(__m128i __V1, __m128i __V2)
405{
406  return __builtin_ia32_pcmpgtq((__v2di)__V1, (__v2di)__V2);
407}
408
409/* SSE4.2 Accumulate CRC32.  */
410static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
411_mm_crc32_u8(unsigned int __C, unsigned char __D)
412{
413  return __builtin_ia32_crc32qi(__C, __D);
414}
415
416static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
417_mm_crc32_u16(unsigned int __C, unsigned short __D)
418{
419  return __builtin_ia32_crc32hi(__C, __D);
420}
421
422static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__))
423_mm_crc32_u32(unsigned int __C, unsigned int __D)
424{
425  return __builtin_ia32_crc32si(__C, __D);
426}
427
428#ifdef __x86_64__
429static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__))
430_mm_crc32_u64(unsigned long long __C, unsigned long long __D)
431{
432  return __builtin_ia32_crc32di(__C, __D);
433}
434#endif /* __x86_64__ */
435
436/* SSE4.2 Population Count.  */
437static __inline__ int __attribute__((__always_inline__, __nodebug__))
438_mm_popcnt_u32(unsigned int __A)
439{
440  return __builtin_popcount(__A);
441}
442
443#ifdef __x86_64__
444static __inline__ long long __attribute__((__always_inline__, __nodebug__))
445_mm_popcnt_u64(unsigned long long __A)
446{
447  return __builtin_popcountll(__A);
448}
449#endif /* __x86_64__ */
450
451#endif /* __SSE4_2__ */
452#endif /* __SSE4_1__ */
453
454#endif /* _SMMINTRIN_H */
455