From 08abd456b8c065b842d3d8de68dbc78ff690a4f2 Mon Sep 17 00:00:00 2001
From: Jonathan Rouzaud-Cornabas <jonathan.rouzaud-cornabas@inria.fr>
Date: Mon, 19 Feb 2018 14:49:28 +0100
Subject: [PATCH] working (but not well) openmp code for SIMD Individual
---
CMakeLists.txt | 2 -
src/libaevol/BitSet_SIMD.h | 21 +++++-
src/libaevol/SIMD_Individual.cpp | 115 +++++++++++++++++++++----------
3 files changed, 99 insertions(+), 39 deletions(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c4abbfd8d..6e7af6043 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,8 +3,6 @@ project(raevol_cuda)
set(__OPENMP_GPU)
-add_definitions(-DWITH_BITSET)
-add_definitions(-D_STATIC_BITSET)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
set(SOURCE_FILES main.cpp src/libaevol/ProxyIndividual.cpp src/libaevol/ProxyIndividual.h src/libaevol/algorithm_cuda.cpp src/libaevol/algorithm_cuda.h src/generate_lookup_map_pow.cpp src/regulation_network.cpp src/ode_solver_only.cpp src/ode_solver_only.h src/ode_solver_only_gpu.h src/ode_solver_only_omp_gpu.cpp src/ode_solver_only_omp_gpu.h src/rk_solver.cpp src/rk_solver.h src/libaevol/CUDA_Individual_cu.h src/libaevol/SIMD_Individual.cpp src/libaevol/SIMD_Individual.h src/libaevol/Dna_SIMD.cpp src/libaevol/Dna_SIMD.h src/libaevol/MutationEvent.cpp src/libaevol/MutationEvent.h src/libaevol/DnaMutator.cpp src/libaevol/DnaMutator.h src/libaevol/Stats_SIMD.cpp src/libaevol/Stats_SIMD.h src/libaevol/BitSet_SIMD.cpp src/libaevol/BitSet_SIMD.h)
diff --git a/src/libaevol/BitSet_SIMD.h b/src/libaevol/BitSet_SIMD.h
index e984dcede..8078268c7 100644
--- a/src/libaevol/BitSet_SIMD.h
+++ b/src/libaevol/BitSet_SIMD.h
@@ -11,6 +11,10 @@
#include <cstdio>
#include <bitset>
+#ifdef _DYNAMIC_BITSET
+#include <boost/dynamic_bitset.hpp>
+#endif
+
namespace aevol {
#define BITSET_BLOCK_SIZE INT32_C(1024)
@@ -78,6 +82,8 @@ class BitSet_SIMD {
data_[(position / CHAR_TO_BITS)] ^= (1 << (position % CHAR_TO_BITS));
#elif _STATIC_BITSET
data_[position].flip();
+#elif _DYNAMIC_BITSET
+ data_[position].flip();
#endif
}
@@ -234,13 +240,24 @@ class BitSet_SIMD {
return dist[0] && dist[1] && dist[2] && dist[3];
}
- __inline bool is_shine_dalgarno_protein_start(bool LEADING, int pos) {
+ __inline bool is_shine_dalgarno_protein_start(bool LEADING, int32_t pos) {
bool start[9] = {false, false, false,
false, false, false,
false, false, false};
#ifdef _STATIC_BITSET
+#pragma omp simd
+ for (int32_t k = 0; k < 9; k++) {
+ int32_t k_t = k >= 6 ? k + 4 : k;
+ if (LEADING)
+ (data_[pos + k_t >= length_ ? pos + k_t - length_ : pos + k_t] ==
+ get(BITSET_SHINE_DAL_SEQ_LEAD,k)) ? (start[k] = true) : (start[k] = false);
+ else
+ (data_[pos - k_t < 0 ? length_ + (pos - k_t) : pos - k_t] ==
+ get(BITSET_SHINE_DAL_SEQ_LAG,k)) ? (start[k] = true) : (start[k] = false);
+ }
+ /*
if (LEADING) {
(data_[pos + 0 >= length_ ? pos + 0 - length_ : pos + 0] ==
get(BITSET_SHINE_DAL_SEQ_LEAD, 0)) ? (start[0] = true)
@@ -299,7 +316,7 @@ class BitSet_SIMD {
(data_[pos - 12 < 0 ? length_ + (pos - 12) : pos - 12] ==
get(BITSET_SHINE_DAL_SEQ_LAG, 8)) ? (start[8] = true)
: (start[8] = false);
- }
+ }*/
#endif
#ifdef _DYNAMIC_CUSTOM_BITSET
diff --git a/src/libaevol/SIMD_Individual.cpp b/src/libaevol/SIMD_Individual.cpp
index d2b10f3d8..708e4e4aa 100644
--- a/src/libaevol/SIMD_Individual.cpp
+++ b/src/libaevol/SIMD_Individual.cpp
@@ -442,8 +442,14 @@ void SIMD_Individual::start_stop_RNA() {
if (indiv_id == 6)
printf("Adding promoters %d at %d\n",dna_pos,prom_idx);*/
- internal_simd_struct[indiv_id]->promoters[prom_idx] = nprom;
- internal_simd_struct[indiv_id]->leading_prom_pos[dna_pos] = prom_idx;
+#pragma omp critical(promoters)
+ {
+ internal_simd_struct[indiv_id]->promoters[prom_idx] = nprom;
+ }
+#pragma omp critical(add_to_promoters_lead)
+ {
+ internal_simd_struct[indiv_id]->leading_prom_pos[dna_pos] = prom_idx;
+ }
}
}
@@ -506,8 +512,14 @@ void SIMD_Individual::start_stop_RNA() {
internal_simd_struct[indiv_id]->count_prom + 1;
}
- internal_simd_struct[indiv_id]->promoters[prom_idx] = nprom;
- internal_simd_struct[indiv_id]->lagging_prom_pos[dna_pos] = prom_idx;
+#pragma omp critical(promoters)
+ {
+ internal_simd_struct[indiv_id]->promoters[prom_idx] = nprom;
+ }
+#pragma omp critical(add_to_promoters_lags)
+ {
+ internal_simd_struct[indiv_id]->lagging_prom_pos[dna_pos] = prom_idx;
+ }
}
}
@@ -538,14 +550,26 @@ void SIMD_Individual::start_stop_RNA() {
void SIMD_Individual::opt_prom_compute_RNA() {
int nb_indiv = exp_m_->nb_indivs();
-#pragma omp parallel
-#pragma omp single
+
+#pragma omp barrier
+
for (int indiv_id = 0; indiv_id < nb_indiv; indiv_id++) {
if (exp_m_->dna_mutator_array_[indiv_id]->hasMutate()) {
internal_simd_struct[indiv_id]->proteins.clear();
internal_simd_struct[indiv_id]->rnas.clear();
internal_simd_struct[indiv_id]->terminator_lead.clear();
internal_simd_struct[indiv_id]->terminator_lag.clear();
+ }
+ }
+
+#pragma omp parallel
+#pragma omp single
+ for (int indiv_id = 0; indiv_id < nb_indiv; indiv_id++) {
+ if (exp_m_->dna_mutator_array_[indiv_id]->hasMutate()) {
+ /*internal_simd_struct[indiv_id]->proteins.clear();
+ internal_simd_struct[indiv_id]->rnas.clear();
+ internal_simd_struct[indiv_id]->terminator_lead.clear();
+ internal_simd_struct[indiv_id]->terminator_lag.clear();*/
for (int rna_idx = 0; rna_idx <
(int) internal_simd_struct[indiv_id]->promoters.size();
@@ -553,13 +577,29 @@ void SIMD_Individual::opt_prom_compute_RNA() {
#pragma omp task firstprivate(indiv_id, rna_idx)
{
- if (internal_simd_struct[indiv_id]->promoters[rna_idx] != nullptr) {
+ promoterStruct* prom;
+#pragma omp critical(promoters)
+ {
+ prom = internal_simd_struct[indiv_id]->promoters[rna_idx];
+ }
+
+ if (prom != nullptr) {
+ int prom_pos;
+ bool lead_lag;
+ double prom_error;
+#pragma omp critical(promoters)
+ {
+ prom_pos = internal_simd_struct[indiv_id]->promoters[rna_idx]->pos;
+ lead_lag = internal_simd_struct[indiv_id]->promoters[rna_idx]->leading_or_lagging;
+ prom_error = fabs(
+ ((float) internal_simd_struct[indiv_id]->promoters[rna_idx]->error));
+ }
- if (internal_simd_struct[indiv_id]->promoters[rna_idx]->leading_or_lagging) {
+ if (lead_lag) {
// if (indiv_id == 152) printf("Searching for RNA (OPT) for indiv %d RNA %d LEAD\n",indiv_id,rna_idx);
/* Search for terminators */
int cur_pos =
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos + 22;
+ prom_pos + 22;
cur_pos = cur_pos >= dna_size[indiv_id] ? cur_pos -
dna_size[indiv_id] :
cur_pos;
@@ -579,8 +619,8 @@ void SIMD_Individual::opt_prom_compute_RNA() {
if (is_term)
#else
+ #pragma omp simd aligned(internal_simd_struct[indiv_id]->dna_->data_:64)
for (int t_motif_id = 0; t_motif_id < 4; t_motif_id++)
- // LEADING
term_dist_leading +=
internal_simd_struct[indiv_id]->dna_->data_[cur_pos + t_motif_id >= dna_size[indiv_id] ? cur_pos +
t_motif_id -
@@ -627,34 +667,34 @@ void SIMD_Individual::opt_prom_compute_RNA() {
}*/
int32_t rna_length = 0;
- if (internal_simd_struct[indiv_id]->promoters[rna_idx]->pos
+ if (prom_pos
> rna_end)
rna_length = dna_size[indiv_id] -
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos
+ prom_pos
+ rna_end;
else
- rna_length = rna_end - internal_simd_struct[indiv_id]->
- promoters[rna_idx]->pos;
+ rna_length = rna_end - prom_pos;
rna_length -= 21;
if (rna_length > 0) {
#pragma omp critical(rnas)
- internal_simd_struct[indiv_id]->rnas.emplace_back(
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos,
- rna_end,
- !internal_simd_struct[indiv_id]->promoters[rna_idx]->leading_or_lagging,
- 1.0 -
- fabs(
- ((float) internal_simd_struct[indiv_id]->promoters[rna_idx]->error)) /
- 5.0, rna_length);
+ {
+ internal_simd_struct[indiv_id]->rnas.emplace_back(
+ prom_pos,
+ rna_end,
+ !lead_lag,
+ 1.0 -
+ prom_error /
+ 5.0, rna_length);
+ }
}
}
// if (indiv_id == 152) printf("Hop to next\n");
} else {
/* Search for terminator */
int cur_pos =
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos - 22;
+ prom_pos - 22;
cur_pos = cur_pos < 0 ? dna_size[indiv_id] + (cur_pos) : cur_pos;
int start_pos = cur_pos;
bool terminator_found = false;
@@ -671,7 +711,8 @@ void SIMD_Individual::opt_prom_compute_RNA() {
if (is_term)
#else
- for (int t_motif_id = 0; t_motif_id < 4; t_motif_id++) {
+ #pragma omp simd aligned(internal_simd_struct[indiv_id]->dna_->data_:64)
+ for (int t_motif_id = 0; t_motif_id < 4; t_motif_id++) {
term_dist_lagging +=
internal_simd_struct[indiv_id]->dna_->data_[cur_pos - t_motif_id < 0 ? cur_pos -
t_motif_id +
@@ -730,14 +771,14 @@ void SIMD_Individual::opt_prom_compute_RNA() {
}*/
int32_t rna_length = 0;
- if (internal_simd_struct[indiv_id]->promoters[rna_idx]->pos <
+ if (prom_pos <
rna_end)
rna_length =
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos +
+ prom_pos +
dna_size[indiv_id] - rna_end;
else
rna_length =
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos -
+ prom_pos -
rna_end;
rna_length -= 21;
@@ -745,14 +786,15 @@ void SIMD_Individual::opt_prom_compute_RNA() {
if (rna_length >= 0) {
#pragma omp critical(rnas)
- internal_simd_struct[indiv_id]->rnas.emplace_back(
- internal_simd_struct[indiv_id]->promoters[rna_idx]->pos,
- rna_end,
- !internal_simd_struct[indiv_id]->promoters[rna_idx]->leading_or_lagging,
- 1.0 -
- fabs(
- ((float) internal_simd_struct[indiv_id]->promoters[rna_idx]->error)) /
- 5.0, rna_length);
+ {
+ internal_simd_struct[indiv_id]->rnas.emplace_back(
+ prom_pos,
+ rna_end,
+ !lead_lag,
+ 1.0 -
+ prom_error /
+ 5.0, rna_length);
+ }
}
// if (indiv_id == 152) printf("Hop to next\n");
@@ -970,6 +1012,8 @@ void SIMD_Individual::start_protein() {
start = internal_simd_struct[indiv_id]->dna_->bitset_->is_shine_dalgarno_protein_start(
true, c_pos);
#else
+
+ #pragma omp simd aligned(internal_simd_struct[indiv_id]->dna_->data_,SHINE_DAL_SEQ_LEAD:64)
for (int k = 0; k < 9; k++) {
k_t = k >= 6 ? k + 4 : k;
t_pos = c_pos + k_t >= dna_size[indiv_id] ? c_pos + k_t -
@@ -992,6 +1036,7 @@ void SIMD_Individual::start_protein() {
start = internal_simd_struct[indiv_id]->dna_->bitset_->is_shine_dalgarno_protein_start(
false, c_pos);
#else
+#pragma omp simd aligned(internal_simd_struct[indiv_id]->dna_->data_,SHINE_DAL_SEQ_LAG:64)
for (int k = 0; k < 9; k++) {
k_t = k >= 6 ? k + 4 : k;
t_pos =
--
GitLab