1  /* { dg-do compile } */
       2  /* { dg-options "-O2 -march=skylake-avx512 -fdump-tree-optimized" } */
       3  /* { dg-final { scan-tree-dump-times ".COND_FMA" 3 "optimized" } } */
       4  /* { dg-final { scan-tree-dump-times ".COND_FNMA" 3 "optimized" } } */
       5  /* { dg-final { scan-tree-dump-times ".COND_FMS" 3 "optimized" } } */
       6  /* { dg-final { scan-tree-dump-times ".COND_FNMS" 3 "optimized" } } */
       7  /* { dg-final { scan-assembler-times "vfmadd132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)"  1 } } */
       8  /* { dg-final { scan-assembler-times "vfnmadd132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)"  1 } } */
       9  /* { dg-final { scan-assembler-times "vfmsub132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)"  1 } } */
      10  /* { dg-final { scan-assembler-times "vfnmsub132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)"  1 } } */
      11  /* { dg-final { scan-assembler-times "vfmadd231pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      12  /* { dg-final { scan-assembler-times "vfnmadd231pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      13  /* { dg-final { scan-assembler-times "vfmsub231pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      14  /* { dg-final { scan-assembler-times "vfnmsub231pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      15  /* { dg-final { scan-assembler-times "vfmadd132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      16  /* { dg-final { scan-assembler-times "vfnmadd132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      17  /* { dg-final { scan-assembler-times "vfmsub132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      18  /* { dg-final { scan-assembler-times "vfnmsub132pd\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)"  1 } } */
      19  
      20  #ifndef NUM
      21  #define NUM 800
      22  #endif
      23  #ifndef TYPE
      24  #define TYPE double
      25  #endif
      26  #ifndef __BUILTIN_FMA
      27  #define __BUILTIN_FMA __builtin_fma
      28  #endif
      29  
      30  TYPE a[NUM], b[NUM], c[NUM], d[NUM], e[NUM], j[NUM];
      31  #define MIN(X,Y) ((X) < (Y) ? (X) : (Y))
      32  #define MAX(X,Y) ((X) < (Y) ? (Y) : (X))
      33  
      34  #define FMA3(OPNAME, OP1, OP2)					\
      35    void								\
      36    __attribute__ ((noipa,optimize ("O3")))			\
      37    foo3_##OPNAME ()						\
      38    {								\
      39      for (int i = 0; i != NUM; i++)				\
      40        {								\
      41  	TYPE tmp = MAX(d[i], e[i]);				\
      42  	if (b[i] < c[i])					\
      43  	  a[i] = __BUILTIN_FMA (OP1 d[i], e[i], OP2 tmp);	\
      44  	else							\
      45  	  a[i] = tmp;						\
      46        }								\
      47    }
      48  
      49  #define FMAZ(OPNAME, OP1, OP2)					\
      50    void								\
      51    __attribute__ ((noipa,optimize ("O3")))			\
      52    fooz_##OPNAME ()						\
      53    {								\
      54      for (int i = 0; i != NUM; i++)				\
      55        if (b[i] < c[i])						\
      56  	a[i] = __BUILTIN_FMA (OP1 d[i], e[i], OP2 a[i]);	\
      57        else							\
      58  	a[i] = .0;						\
      59    }
      60  
      61  #define FMA1(OPNAME, OP1, OP2)					\
      62    void								\
      63    __attribute__ ((noipa,optimize ("O3")))			\
      64    foo1_##OPNAME ()						\
      65    {								\
      66      for (int i = 0; i != NUM; i++)				\
      67        if (b[i] < c[i])						\
      68  	a[i] = __BUILTIN_FMA (OP1 d[i], e[i], OP2 a[i]);	\
      69        else							\
      70  	a[i] = d[i];						\
      71    }
      72  
      73  
      74  FMAZ (fma,, +);
      75  FMAZ (fms,, -);
      76  FMAZ (fnma, -, +);
      77  FMAZ (fnms, -, -);
      78  
      79  FMA1 (fma,, +);
      80  FMA1 (fms,, -);
      81  FMA1 (fnma, -, +);
      82  FMA1 (fnms, -, -);
      83  
      84  FMA3 (fma,, +);
      85  FMA3 (fms,, -);
      86  FMA3 (fnma, -, +);
      87  FMA3 (fnms, -, -);