Skip to content

GitLab

Commit 4d6d1d6b authored by Edgar Solomonik's avatar Edgar Solomonik
Browse files

Got normal, ror, isend/irecv, put, and fompi put all working and integrated in unified code

parent 69aee1c7
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing with 63 additions and 54 deletions
src/contraction/Makefile
View file @ 4d6d1d6b
......@@ -3,7 +3,7 @@ include ../../config.mk
OBJS = contraction.o sym_seq_ctr.o ctr_offload.o ctr_comm.o ctr_tsr.o ctr_2d_general.o
#%d | r ! grep -ho "\.\..*\.h" *.cxx *.h | sort | uniq
HDRS = ../../Makefile ../../config.mk ../interface/functions.h ../mapping/distribution.h ../mapping/mapping.h ../redistribution/nosym_transp.h ../redistribution/redist.h ../scaling/strp_tsr.h ../shared/iter_tsr.h ../shared/memcontrol.h ../shared/offload.h ../shared/util.h ../symmetry/sym_indices.h ../symmetry/symmetrization.h ../tensor/algstrct.h ../tensor/untyped_tensor.h
HDRS = ../../Makefile ../../config.mk ../interface/functions.h ../mapping/distribution.h ../mapping/mapping.h ../redistribution/nosym_transp.h ../redistribution/redist.h ../scaling/strp_tsr.h ../shared/iter_tsr.h ../shared/memcontrol.h ../shared/offload.h ../shared/util.h ../symmetry/sym_indices.h ../symmetry/symmetrization.h ../tensor/algstrct.h ../tensor/untyped_tensor.h
ctf: $(OBJS)
......
src/interface/common.h
View file @ 4d6d1d6b
......@@ -13,6 +13,8 @@
#include <unistd.h>
#include <iostream>
#include <limits.h>
#include "../shared/fompi_wrapper.h"
#include "mpi.h"
/**
......
src/redistribution/phase_reshuffle.cxx
View file @ 4d6d1d6b
......@@ -6,7 +6,7 @@
#include "nosym_transp.h"
#define MTAG 777
#define ROR
//#define ROR
#ifdef ROR
//#define IREDIST
//#define REDIST_PUT
......@@ -149,42 +149,39 @@ namespace CTF_int {
template <int idim>
void calc_cnt_from_rep_cnt(int const * rep_phase,
int const * rep_phase_lda,
int const * rank,
int const * new_pe_lda,
int const * old_phys_phase,
int const * new_phys_phase,
int * const * pe_offset,
int * const * bucket_offset,
int64_t const * old_counts,
int64_t * counts,
int coff,
int roff,
int bucket_off,
int pe_off,
int dir){
for (int i=0; i<rep_phase[idim]; i++){
calc_cnt_from_rep_cnt<idim-1>(rep_phase, rep_phase_lda, rank, new_pe_lda, old_phys_phase, new_phys_phase, old_counts, counts,
coff+new_pe_lda[idim]*((rank[idim]+i*old_phys_phase[idim])%new_phys_phase[idim]),
roff+rep_phase_lda[idim]*i, dir);
int rec_bucket_off = bucket_off+bucket_offset[idim][i];
int rec_pe_off = pe_off+pe_offset[idim][i];
calc_cnt_from_rep_cnt<idim-1>(rep_phase, pe_offset, bucket_offset, old_counts, counts, rec_bucket_off, rec_pe_off, dir);
// coff+new_pe_lda[idim]*((rank[idim]+i*old_phys_phase[idim])%new_phys_phase[idim]),
// roff+rep_phase_lda[idim]*i, dir);
}
}
template <>
void calc_cnt_from_rep_cnt<0>(int const * rep_phase,
int const * rep_phase_lda,
int const * rank,
int const * new_pe_lda,
int const * old_phys_phase,
int const * new_phys_phase,
int64_t const * old_counts,
int64_t * counts,
int coff,
int roff,
int dir){
void calc_cnt_from_rep_cnt<0>
(int const * rep_phase,
int * const * pe_offset,
int * const * bucket_offset,
int64_t const * old_counts,
int64_t * counts,
int bucket_off,
int pe_off,
int dir){
if (dir){
for (int i=0; i<rep_phase[0]; i++){
counts[coff+new_pe_lda[0]*((rank[0]+i*old_phys_phase[0])%new_phys_phase[0])] = old_counts[roff + i];
counts[pe_off+pe_offset[0][i]] = old_counts[bucket_off+i];
}
} else {
for (int i=0; i<rep_phase[0]; i++){
counts[roff + i] = old_counts[coff+new_pe_lda[0]*((rank[0]+i*old_phys_phase[0])%new_phys_phase[0])];
counts[bucket_off+i] = old_counts[pe_off+pe_offset[0][i]];
}
}
......@@ -209,6 +206,7 @@ namespace CTF_int {
new_loc_edge_len = (int*)alloc(order*sizeof(int));
int nrep = 1;
for (int i=0; i<order; i++){
//FIXME: computed elsewhere already
rep_phase_lda[i] = nrep;
sphase[i] = lcm(old_dist.phys_phase[i],new_dist.phys_phase[i]);
rep_phase[i] = sphase[i] / old_dist.phys_phase[i];
......@@ -436,9 +434,9 @@ namespace CTF_int {
MPI_Comm cm,
char * buffer,
algstrct const * sr,
int bucket_off=0,
int pe_off=0,
int dir=0){
int bucket_off,
int pe_off,
int dir){
for (int r=0; r<rep_phase[idim]; r++){
int rec_bucket_off = bucket_off+bucket_offset[idim][r];
int rec_pe_off = pe_off+pe_offset[idim][r];
......@@ -461,7 +459,7 @@ namespace CTF_int {
int pe_off,
int dir){
for (int r=0; r<rep_phase[0]; r++){
int bucket = bucket_off+bucket_offset[0][r];
int bucket = bucket_off+r;
int pe = pe_off+pe_offset[0][r];
if (dir)
MPI_Irecv(buffer+displs[bucket]*sr->el_size, counts[bucket], sr->mdtype(), pe, MTAG, cm, reqs+bucket);
......@@ -539,7 +537,7 @@ namespace CTF_int {
for (int r=0; r<rep_phase[idim]; r++){
int rec_bucket_off = bucket_off+bucket_offset[idim][r];
int rec_pe_off = pe_off+pe_offset[idim][r];
put_buckets<idim-1>(rep_phase, bucket_offset, buckets, counts, sr, put_displs, win, rec_bucket_off, rec_pe_off);
put_buckets<idim-1>(rep_phase, pe_offset, bucket_offset, buckets, counts, sr, put_displs, win, rec_bucket_off, rec_pe_off);
}
}
......@@ -802,15 +800,19 @@ namespace CTF_int {
int nold_rep = 1;
int * old_rep_phase; alloc_ptr(sizeof(int)*order, (void**)&old_rep_phase);
int * old_rep_phase_lda; alloc_ptr(sizeof(int)*order, (void**)&old_rep_phase_lda);
for (int i=0; i<order; i++){
old_rep_phase[i] = lcm(old_dist.phys_phase[i], new_dist.phys_phase[i])/old_dist.phys_phase[i];
old_rep_phase_lda[i] = nold_rep;
nold_rep *= old_rep_phase[i];
}
int nnew_rep = 1;
int * new_rep_phase; alloc_ptr(sizeof(int)*order, (void**)&new_rep_phase);
int * new_rep_phase_lda; alloc_ptr(sizeof(int)*order, (void**)&new_rep_phase_lda);
for (int i=0; i<order; i++){
new_rep_phase[i] = lcm(new_dist.phys_phase[i], old_dist.phys_phase[i])/new_dist.phys_phase[i];
new_rep_phase_lda[i] = nnew_rep;
nnew_rep *= new_rep_phase[i];
}
......@@ -848,10 +850,9 @@ namespace CTF_int {
precompute_offsets(old_dist, new_dist, sym, edge_len, old_rep_phase, old_phys_edge_len, old_virt_edge_len, old_dist.virt_phase, old_virt_lda, old_virt_nelem, send_pe_offset, send_bucket_offset, send_data_offset);
#ifdef IREDIST
if (new_idx_lyr == 0)
SWITCH_ORD_CALL(isendrecv, order-1, recv_pe_offset, recv_bucket_offset, new_rep_phase, recv_counts, recv_displs, recv_reqs, ord_glb_comm.cm, recv_buffer, sr);
SWITCH_ORD_CALL(isendrecv, order-1, recv_pe_offset, recv_bucket_offset, new_rep_phase, recv_counts, recv_displs, recv_reqs, ord_glb_comm.cm, recv_buffer, sr, 0, 0, 1);
#endif
#ifndef IREDIST
#ifndef REDIST_PUT
......@@ -861,16 +862,15 @@ namespace CTF_int {
send_displs[i] = send_displs[i-1] + send_counts[i-1];
}
#else
int64_t * all_get_displs = (int64_t*)alloc(sizeof(int64_t)*ord_glb_comm.np);
SWITCH_ORD_CALL(calc_cnt_from_rep_cnt, order-1, rep_phase, rep_phase_lda, old_dist.perank, new_dist.pe_lda, old_dist.phys_phase, new_dist.phys_phase, counts, all_get_displs);
int64_t * all_recv_displs = (int64_t*)alloc(sizeof(int64_t)*ord_glb_comm.np);
SWITCH_ORD_CALL(calc_cnt_from_rep_cnt, order-1, new_rep_phase, recv_pe_offset, recv_bucket_offset, recv_displs, all_recv_displs, 0, 0, 1);
int64_t * all_put_displs = (int64_t*)alloc(sizeof(int64_t)*ord_glb_comm.np);
MPI_Alltoall(all_get_displs, 1, MPI_INT64_T, all_put_displs, 1, MPI_INT64_T, ord_glb_comm.cm);
cfree(all_get_displs);
MPI_Alltoall(all_recv_displs, 1, MPI_INT64_T, all_put_displs, 1, MPI_INT64_T, ord_glb_comm.cm);
cfree(all_recv_displs);
int64_t * put_displs = (int64_t*)alloc(sizeof(int64_t)*nold_rep);
SWITCH_ORD_CALL(calc_cnt_from_rep_cnt, order-1, rep_phase, rep_phase_lda, new_dist.perank, old_dist.pe_lda, new_dist.phys_phase, old_dist.phys_phase, all_put_displs, put_displs);
SWITCH_ORD_CALL(calc_cnt_from_rep_cnt, order-1, old_rep_phase, send_pe_offset, send_bucket_offset, all_put_displs, put_displs, 0, 0, 0);
cfree(all_put_displs);
......@@ -1001,6 +1001,9 @@ namespace CTF_int {
new_rep_phase, new_rep_idx, new_dist.virt_phase[0],
#ifdef IREDIST
recv_reqs, ord_glb_comm.cm,
#endif
#ifdef REDIST_PUT
NULL, win,
#endif
0, aux_buf, buckets, recv_counts, sr);
cfree(new_rep_idx);
......
src/redistribution/phase_reshuffle.h
View file @ 4d6d1d6b
......@@ -51,19 +51,6 @@ namespace CTF_int {
int const * edge_len,
int const * loc_edge_len);
template <int idim>
void calc_cnt_from_rep_cnt(int const * rep_phase,
int const * rep_phase_lda,
int const * rank,
int const * new_pe_lda,
int const * old_phys_phase,
int const * new_phys_phase,
int64_t const * old_counts,
int64_t * counts,
int coff=0,
int roff=0,
int dir=1);
void calc_drv_displs(int const * sym,
int const * edge_len,
int const * loc_edge_len,
......
src/shared/fompi_wrapper.h 0 → 100644
View file @ 4d6d1d6b
#ifndef __FOMPI_WRAPPER__
#define __FOMPI_WRAPPER__
#ifdef USE_FOMPI
#include "fompi.h"
#define MPI_Init(...) foMPI_Init(__VA_ARGS__)
#define MPI_Win(...) foMPI_Win(__VA_ARGS__)
#define MPI_Win_create(...) foMPI_Win_create(__VA_ARGS__)
#define MPI_Win_fence(...) foMPI_Win_fence(__VA_ARGS__)
#define MPI_Win_free(...) foMPI_Win_free(__VA_ARGS__)
#define MPI_Put(...) foMPI_Put(__VA_ARGS__)
#endif
#endif
src/shared/util.h
View file @ 4d6d1d6b
......@@ -28,7 +28,7 @@ namespace CTF_int {
/* Force redistributions always by setting to 1 */
#define REDIST 0
//#define VERIFY 0
#define VERIFY_REMAP 0
#define VERIFY_REMAP 1
#define FOLD_TSR 1
#define PERFORM_DESYM 1
#define ALLOW_NVIRT 1024
......
test/test_suite.cxx
View file @ 4d6d1d6b
......@@ -41,8 +41,9 @@ int main(int argc, char ** argv){
int in_num = argc;
char ** input_str = argv;
int nt;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &nt);
//int nt;
//MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &nt);
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &np);
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment