Chris@82: /*
Chris@82:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@82:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@82:  * Copyright (c) 2012-2013 Romain Dolbeau
Chris@82:  *
Chris@82:  * Permission is hereby granted, free of charge, to any person obtaining a copy
Chris@82:  * of this software and associated documentation files (the "Software"), to deal
Chris@82:  * in the Software without restriction, including without limitation the rights
Chris@82:  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
Chris@82:  * copies of the Software, and to permit persons to whom the Software is
Chris@82:  * furnished to do so, subject to the following conditions:
Chris@82:  *
Chris@82:  * The above copyright notice and this permission notice shall be included in
Chris@82:  * all copies or substantial portions of the Software.
Chris@82:  *
Chris@82:  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
Chris@82:  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
Chris@82:  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
Chris@82:  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
Chris@82:  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
Chris@82:  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
Chris@82:  * THE SOFTWARE.
Chris@82:  *
Chris@82:  */
Chris@82: 
Chris@82: #include "kernel/ifftw.h"
Chris@82: 
Chris@82: #if HAVE_AVX512
Chris@82: 
Chris@82: #if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
Chris@82: 
Chris@82: #include "amd64-cpuid.h"
Chris@82: 
Chris@82: int X(have_simd_avx512)(void)
Chris@82: {
Chris@82:      static int init = 0, res;     
Chris@82:      int max_stdfn, eax, ebx, ecx, edx;
Chris@82: 
Chris@82:      /* NOTE: this code is a total guess.  I don't have an avx512
Chris@82:         machine available.  The code contributed by Erik Lindahl would
Chris@82:         crash on a machine without XGETBV, so I had to guess a fix. */
Chris@82:      if (!init) {
Chris@82:           cpuid_all(0,0,&eax,&ebx,&ecx,&edx);
Chris@82:           max_stdfn = eax;
Chris@82:           if (max_stdfn >= 0x1) {
Chris@82:                /* have OSXSAVE? (implies XGETBV exists) */
Chris@82:                cpuid_all(0x1, 0, &eax, &ebx, &ecx, &edx);
Chris@82:                if ((ecx & 0x08000000) == 0x08000000) {
Chris@82:                     /* have AVX512? */
Chris@82:                     cpuid_all(7,0,&eax,&ebx,&ecx,&edx);
Chris@82:                     if (ebx & (1 << 16)) {
Chris@82:                          /* have OS support for XMM, YMM, ZMM */
Chris@82:                          int zmm_ymm_xmm = (7 << 5) | (1 << 2) | (1 << 1);
Chris@82:                          res = ((xgetbv_eax(0) & zmm_ymm_xmm) == zmm_ymm_xmm);
Chris@82:                     }
Chris@82:                }
Chris@82:           }
Chris@82:           init = 1;
Chris@82:      }
Chris@82: 
Chris@82:      return res;
Chris@82: }
Chris@82: 
Chris@82: #else /* 32-bit code */
Chris@82: 
Chris@82: #error "Avx512 is 64 bits only"
Chris@82: 
Chris@82: #endif
Chris@82: 
Chris@82: #endif