void
gmx_setup_kernels(FILE *fplog,gmx_bool bGenericKernelOnly)
{
int i;
snew(nb_kernel_list,eNR_NBKERNEL_NR);
snew(pf_nb_kernel_list,eNR_NBKERNEL_NR);
/* Note that later calls overwrite earlier, so the preferred (fastest)
* version should be at the end. For instance, we call SSE after 3DNow.
*/
for(i=0; i<eNR_NBKERNEL_NR; i++)
{
nb_kernel_list[i] = NULL;
pf_nb_kernel_list[i] = NULL;
}
if (bGenericKernelOnly)
{
return;
}
if(fplog)
{
fprintf(fplog,"Configuring nonbonded kernels...\n");
}
nb_kernel_setup(fplog,nb_kernel_list);
if(getenv("GMX_NOOPTIMIZEDKERNELS") != NULL)
{
return;
}
/* Setup kernels. The last called setup routine will overwrite earlier assignments,
* so we should e.g. test SSE3 support _after_ SSE2 support,
* and call e.g. Fortran setup before SSE.
*/
/* Setup the pairwise-force kernels, c/fortran only, no Assembly loops available */
if(fplog)
fprintf(fplog, "Configuring pairwise force nonbonded kernels...\n");
pf_nb_kernel_setup(fplog,pf_nb_kernel_list);
#if defined(GMX_FORTRAN) && defined(GMX_DOUBLE)
nb_kernel_setup_f77_double(fplog,nb_kernel_list);
#endif
#if defined(GMX_FORTRAN) && !defined(GMX_DOUBLE)
nb_kernel_setup_f77_single(fplog,nb_kernel_list);
#endif
#ifdef GMX_BLUEGENE
nb_kernel_setup_bluegene(fplog,nb_kernel_list);
#endif
#ifdef GMX_POWER6
nb_kernel_setup_power6(fplog,nb_kernel_list);
#endif
#ifdef GMX_PPC_ALTIVEC
nb_kernel_setup_ppc_altivec(fplog,nb_kernel_list);
#endif
#if defined(GMX_IA32_SSE)
nb_kernel_setup_ia32_sse(fplog,nb_kernel_list);
#endif
#if defined(GMX_IA32_SSE2)
nb_kernel_setup_ia32_sse2(fplog,nb_kernel_list);
#endif
#if defined(GMX_X86_64_SSE)
nb_kernel_setup_x86_64_sse(fplog,nb_kernel_list);
#endif
#if defined(GMX_X86_64_SSE2)
nb_kernel_setup_x86_64_sse2(fplog,nb_kernel_list);
#endif
#if (defined GMX_IA64_ASM && defined GMX_DOUBLE)
nb_kernel_setup_ia64_double(fplog,nb_kernel_list);
#endif
#if (defined GMX_IA64_ASM && !defined GMX_DOUBLE)
nb_kernel_setup_ia64_single(fplog,nb_kernel_list);
#endif
if(fplog)
{
fprintf(fplog,"\n\n");
}
}
void
gmx_setup_kernels(FILE *fplog)
{
int i;
snew(nb_kernel_list,eNR_NBKERNEL_NR);
/* Note that later calls overwrite earlier, so the preferred (fastest)
* version should be at the end. For instance, we call SSE after 3DNow.
*/
for(i=0; i<eNR_NBKERNEL_NR; i++)
{
nb_kernel_list[i] = NULL;
}
if(getenv("GMX_NB_GENERIC") != NULL)
{
if(fplog)
{
fprintf(fplog,
"Found environment variable GMX_NB_GENERIC.\n"
"Disabling all interaction-specific nonbonded kernels.\n\n");
}
return;
}
if(fplog)
{
fprintf(fplog,"Configuring nonbonded kernels...\n");
}
nb_kernel_setup(fplog,nb_kernel_list);
if(getenv("GMX_NOOPTIMIZEDKERNELS") != NULL)
{
if(fplog)
fprintf(fplog,
"Found environment variable GMX_NOOPTIMIZEDKERNELS.\n"
"Disabling all SSE/SSE2/Altivec/ia64/Power6/Bluegene specific kernels.\n\n");
return;
}
/* Setup kernels. The last called setup routine will overwrite earlier assignments,
* so we should e.g. test SSE3 support _after_ SSE2 support.
*/
#ifdef GMX_PPC_ALTIVEC
nb_kernel_setup_ppc_altivec(fplog,nb_kernel_list);
#endif
#if defined(GMX_IA32_SSE)
nb_kernel_setup_ia32_sse(fplog,nb_kernel_list);
#endif
#if defined(GMX_IA32_SSE2)
nb_kernel_setup_ia32_sse2(fplog,nb_kernel_list);
#endif
#if defined(GMX_X86_64_SSE)
nb_kernel_setup_x86_64_sse(fplog,nb_kernel_list);
#endif
#if defined(GMX_X86_64_SSE2)
nb_kernel_setup_x86_64_sse2(fplog,nb_kernel_list);
#endif
#if defined(GMX_SSE2)
# ifdef GMX_DOUBLE
nb_kernel_setup_sse2_double(fplog,nb_kernel_list);
# else
nb_kernel_setup_sse2_single(fplog,nb_kernel_list);
# endif
#endif
#if defined(GMX_FORTRAN) && defined(GMX_DOUBLE)
nb_kernel_setup_f77_double(fplog,nb_kernel_list);
#endif
#if defined(GMX_FORTRAN) && !defined(GMX_DOUBLE)
nb_kernel_setup_f77_single(fplog,nb_kernel_list);
#endif
#if (defined GMX_IA64_ASM && defined GMX_DOUBLE)
nb_kernel_setup_ia64_double(fplog,nb_kernel_list);
#endif
#if (defined GMX_IA64_ASM && !defined GMX_DOUBLE)
nb_kernel_setup_ia64_single(fplog,nb_kernel_list);
#endif
#ifdef GMX_BLUEGENE
nb_kernel_setup_bluegene(fplog,nb_kernel_list);
#endif
#ifdef GMX_POWER6
nb_kernel_setup_power6(fplog,nb_kernel_list);
#endif
if(fplog)
{
fprintf(fplog,"\n\n");
}
}