Chris@69: /* Copyright (c) 2014, Cisco Systems, INC
Chris@69:    Written by XiangMingZhu WeiZhou MinPeng YanWang
Chris@69: 
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: #if !defined(X86CPU_H)
Chris@69: # define X86CPU_H
Chris@69: 
Chris@69: # if defined(OPUS_X86_MAY_HAVE_SSE)
Chris@69: #  define MAY_HAVE_SSE(name) name ## _sse
Chris@69: # else
Chris@69: #  define MAY_HAVE_SSE(name) name ## _c
Chris@69: # endif
Chris@69: 
Chris@69: # if defined(OPUS_X86_MAY_HAVE_SSE2)
Chris@69: #  define MAY_HAVE_SSE2(name) name ## _sse2
Chris@69: # else
Chris@69: #  define MAY_HAVE_SSE2(name) name ## _c
Chris@69: # endif
Chris@69: 
Chris@69: # if defined(OPUS_X86_MAY_HAVE_SSE4_1)
Chris@69: #  define MAY_HAVE_SSE4_1(name) name ## _sse4_1
Chris@69: # else
Chris@69: #  define MAY_HAVE_SSE4_1(name) name ## _c
Chris@69: # endif
Chris@69: 
Chris@69: # if defined(OPUS_X86_MAY_HAVE_AVX)
Chris@69: #  define MAY_HAVE_AVX(name) name ## _avx
Chris@69: # else
Chris@69: #  define MAY_HAVE_AVX(name) name ## _c
Chris@69: # endif
Chris@69: 
Chris@69: # if defined(OPUS_HAVE_RTCD)
Chris@69: int opus_select_arch(void);
Chris@69: # endif
Chris@69: 
Chris@69: /*gcc appears to emit MOVDQA's to load the argument of an _mm_cvtepi8_epi32()
Chris@69:   or _mm_cvtepi16_epi32() when optimizations are disabled, even though the
Chris@69:   actual PMOVSXWD instruction takes an m32 or m64. Unlike a normal memory
Chris@69:   reference, these require 16-byte alignment and load a full 16 bytes (instead
Chris@69:   of 4 or 8), possibly reading out of bounds.
Chris@69: 
Chris@69:   We can insert an explicit MOVD or MOVQ using _mm_cvtsi32_si128() or
Chris@69:   _mm_loadl_epi64(), which should have the same semantics as an m32 or m64
Chris@69:   reference in the PMOVSXWD instruction itself, but gcc is not smart enough to
Chris@69:   optimize this out when optimizations ARE enabled.
Chris@69: 
Chris@69:   Clang, in contrast, requires us to do this always for _mm_cvtepi8_epi32
Chris@69:   (which is fair, since technically the compiler is always allowed to do the
Chris@69:   dereference before invoking the function implementing the intrinsic).
Chris@69:   However, it is smart enough to eliminate the extra MOVD instruction.
Chris@69:   For _mm_cvtepi16_epi32, it does the right thing, though does *not* optimize out
Chris@69:   the extra MOVQ if it's specified explicitly */
Chris@69: 
Chris@69: # if defined(__clang__) || !defined(__OPTIMIZE__)
Chris@69: #  define OP_CVTEPI8_EPI32_M32(x) \
Chris@69:  (_mm_cvtepi8_epi32(_mm_cvtsi32_si128(*(int *)(x))))
Chris@69: # else
Chris@69: #  define OP_CVTEPI8_EPI32_M32(x) \
Chris@69:  (_mm_cvtepi8_epi32(*(__m128i *)(x)))
Chris@69: #endif
Chris@69: 
Chris@69: /* similar reasoning about the instruction sequence as in the 32-bit macro above,
Chris@69:  */
Chris@69: # if defined(__clang__) || !defined(__OPTIMIZE__)
Chris@69: #  define OP_CVTEPI16_EPI32_M64(x) \
Chris@69:  (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x))))
Chris@69: # else
Chris@69: #  define OP_CVTEPI16_EPI32_M64(x) \
Chris@69:  (_mm_cvtepi16_epi32(*(__m128i *)(x)))
Chris@69: # endif
Chris@69: 
Chris@69: #endif