bool
ilo_shader_compile_gs_passthrough(const struct ilo_shader_state *vs_state,
const struct ilo_shader_variant *vs_variant,
const int *so_mapping,
struct ilo_shader *vs)
{
struct gs_compile_context gcc;
struct ilo_shader_state state;
struct ilo_shader_variant variant;
const int num_verts = 3;
int i;
/* init GS state and variant */
state = *vs_state;
state.info.tokens = NULL;
for (i = 0; i < state.info.stream_output.num_outputs; i++) {
const int reg = state.info.stream_output.output[i].register_index;
state.info.stream_output.output[i].register_index = so_mapping[reg];
}
variant = *vs_variant;
variant.u.gs.rasterizer_discard = vs_variant->u.vs.rasterizer_discard;
variant.u.gs.num_inputs = vs->out.count;
for (i = 0; i < vs->out.count; i++) {
variant.u.gs.semantic_names[i] =
vs->out.semantic_names[i];
variant.u.gs.semantic_indices[i] =
vs->out.semantic_indices[i];
}
if (!gs_setup(&gcc, &state, &variant, num_verts))
return false;
if (!gs_compile_passthrough(&gcc)) {
FREE(gcc.shader);
gcc.shader = NULL;
}
/* no need to call toy_tgsi_cleanup() */
toy_compiler_cleanup(&gcc.tc);
return append_gs_to_vs(vs, gcc.shader, num_verts);
}
/**
* Compile the geometry shader.
*/
struct ilo_shader *
ilo_shader_compile_gs(const struct ilo_shader_state *state,
const struct ilo_shader_variant *variant)
{
struct gs_compile_context gcc;
if (!gs_setup(&gcc, state, variant, 0))
return NULL;
if (!gs_compile(&gcc)) {
FREE(gcc.shader);
gcc.shader = NULL;
}
toy_tgsi_cleanup(&gcc.tgsi);
toy_compiler_cleanup(&gcc.tc);
return gcc.shader;;
}
int main(int narg, char *arg[])
{
comm_ext world; int np;
double t1,t2,gs_time,mpi_time;
struct gs_data *gsh;
struct comm comm;
int *localData,*recvBuf;
slong *glo_num;
int i,j,nid,nsamples;
T *v;
#ifdef MPI
MPI_Init(&narg,&arg);
world = MPI_COMM_WORLD;
MPI_Comm_size(world,&np);
#else
world=0, np=1;
#endif
comm_init(&comm,world);
MPI_Comm_rank(world,&nid);
glo_num = malloc(sizeof(slong)*np);
for(i=1;i<=np;i++){
j = nid+1;
if(j>=i){
glo_num[i-1] = (i-1)*np-i*(i-1)/2 + j-1;
} else {
glo_num[i-1] = (j-1)*np - j*(j-1)/2+i-1;
}
if(j==i){
glo_num[i-1] = 0;
}
}
gsh = gs_setup(glo_num,np,&comm,0,gs_auto,1);
nsamples = 10000;
localData = malloc(sizeof(int)*np);
recvBuf = malloc(sizeof(int)*np);
MPI_Barrier(world);
t1 = MPI_Wtime();
for(j=0;j<nsamples;j++){
for(i=0;i<np;i++){
localData[i] = nid+i;
recvBuf[i] = nid+i;
}
gs(recvBuf,gs_int,gs_add,0,gsh,0);
for(i=0;i<np;i++){
recvBuf[i] = recvBuf[i] - localData[i];
}
}
MPI_Barrier(world);
t2 = MPI_Wtime();
gs_time = t2 - t1;
MPI_Barrier(world);
t1 = MPI_Wtime();
for(j=0;j<nsamples;j++){
for(i=0;i<np;i++){
localData[i] = nid+i;
}
MPI_Alltoall(localData,1,MPI_INT,recvBuf,1,MPI_INT,world);
}
MPI_Barrier(world);
t2 = MPI_Wtime();
mpi_time = t2 - t1;
if(nid==0)printf("gs_time: %f mpi_time: %f\n",gs_time,mpi_time);
#ifdef MPI
MPI_Barrier(world);
// MPI_Finalize();
#endif
return 0;
}
