Subversion Repositories QNX 8.QNX8 LLVM/Clang compiler suite

/*===---- pmmintrin.h - Implementation of SSE3 intrinsics on PowerPC -------===

 *

 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

 * See https://llvm.org/LICENSE.txt for license information.

 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

 *

 *===-----------------------------------------------------------------------===

 */

/* Implemented from the specification included in the Intel C++ Compiler

   User Guide and Reference, version 9.0.  */

#ifndef NO_WARN_X86_INTRINSICS

/* This header is distributed to simplify porting x86_64 code that

   makes explicit use of Intel intrinsics to powerpc64le.

   It is the user's responsibility to determine if the results are

   acceptable and make additional changes as necessary.

   Note that much code that uses Intel intrinsics can be rewritten in

   standard C or GNU C extensions, which are more portable and better

   optimized across multiple targets.

   In the specific case of X86 SSE3 intrinsics, the PowerPC VMX/VSX ISA

   is a good match for most SIMD operations.  However the Horizontal

   add/sub requires the data pairs be permuted into a separate

   registers with vertical even/odd alignment for the operation.

   And the addsub operation requires the sign of only the even numbered

   elements be flipped (xored with -0.0).

   For larger blocks of code using these intrinsic implementations,

   the compiler be should be able to schedule instructions to avoid

   additional latency.

   In the specific case of the monitor and mwait instructions there are

   no direct equivalent in the PowerISA at this time.  So those

   intrinsics are not implemented.  */

#error                                                                         \

    "Please read comment above.  Use -DNO_WARN_X86_INTRINSICS to disable this warning."

#endif

#ifndef PMMINTRIN_H_

#define PMMINTRIN_H_

#if defined(__powerpc64__) &&                                                  \

    (defined(__linux__) || defined(__FreeBSD__) || defined(_AIX))

/* We need definitions from the SSE2 and SSE header files*/

#include <emmintrin.h>

extern __inline __m128

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_addsub_ps(__m128 __X, __m128 __Y) {

  const __v4sf __even_n0 = {-0.0, 0.0, -0.0, 0.0};

  __v4sf __even_neg_Y = vec_xor(__Y, __even_n0);

  return (__m128)vec_add(__X, __even_neg_Y);

}

extern __inline __m128d

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_addsub_pd(__m128d __X, __m128d __Y) {

  const __v2df __even_n0 = {-0.0, 0.0};

  __v2df __even_neg_Y = vec_xor(__Y, __even_n0);

  return (__m128d)vec_add(__X, __even_neg_Y);

}

extern __inline __m128

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_hadd_ps(__m128 __X, __m128 __Y) {

  __vector unsigned char __xform2 = {0x00, 0x01, 0x02, 0x03, 0x08, 0x09,

                                     0x0A, 0x0B, 0x10, 0x11, 0x12, 0x13,

                                     0x18, 0x19, 0x1A, 0x1B};

  __vector unsigned char __xform1 = {0x04, 0x05, 0x06, 0x07, 0x0C, 0x0D,

                                     0x0E, 0x0F, 0x14, 0x15, 0x16, 0x17,

                                     0x1C, 0x1D, 0x1E, 0x1F};

  return (__m128)vec_add(vec_perm((__v4sf)__X, (__v4sf)__Y, __xform2),

                         vec_perm((__v4sf)__X, (__v4sf)__Y, __xform1));

}

extern __inline __m128

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_hsub_ps(__m128 __X, __m128 __Y) {

  __vector unsigned char __xform2 = {0x00, 0x01, 0x02, 0x03, 0x08, 0x09,

                                     0x0A, 0x0B, 0x10, 0x11, 0x12, 0x13,

                                     0x18, 0x19, 0x1A, 0x1B};

  __vector unsigned char __xform1 = {0x04, 0x05, 0x06, 0x07, 0x0C, 0x0D,

                                     0x0E, 0x0F, 0x14, 0x15, 0x16, 0x17,

                                     0x1C, 0x1D, 0x1E, 0x1F};

  return (__m128)vec_sub(vec_perm((__v4sf)__X, (__v4sf)__Y, __xform2),

                         vec_perm((__v4sf)__X, (__v4sf)__Y, __xform1));

}

extern __inline __m128d

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_hadd_pd(__m128d __X, __m128d __Y) {

  return (__m128d)vec_add(vec_mergeh((__v2df)__X, (__v2df)__Y),

                          vec_mergel((__v2df)__X, (__v2df)__Y));

}

extern __inline __m128d

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_hsub_pd(__m128d __X, __m128d __Y) {

  return (__m128d)vec_sub(vec_mergeh((__v2df)__X, (__v2df)__Y),

                          vec_mergel((__v2df)__X, (__v2df)__Y));

}

#ifdef _ARCH_PWR8

extern __inline __m128

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_movehdup_ps(__m128 __X) {

  return (__m128)vec_mergeo((__v4su)__X, (__v4su)__X);

}

#endif

#ifdef _ARCH_PWR8

extern __inline __m128

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_moveldup_ps(__m128 __X) {

  return (__m128)vec_mergee((__v4su)__X, (__v4su)__X);

}

#endif

extern __inline __m128d

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_loaddup_pd(double const *__P) {

  return (__m128d)vec_splats(*__P);

}

extern __inline __m128d

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_movedup_pd(__m128d __X) {

  return _mm_shuffle_pd(__X, __X, _MM_SHUFFLE2(0, 0));

}

extern __inline __m128i

    __attribute__((__gnu_inline__, __always_inline__, __artificial__))

    _mm_lddqu_si128(__m128i const *__P) {

  return (__m128i)(vec_vsx_ld(0, (signed int const *)__P));

}

/* POWER8 / POWER9 have no equivalent for _mm_monitor nor _mm_wait.  */

#else

#include_next <pmmintrin.h>

#endif /* defined(__powerpc64__) &&                                            \

        *   (defined(__linux__) || defined(__FreeBSD__) || defined(_AIX)) */

#endif /* PMMINTRIN_H_ */