From: Ian Jackson <ian@davenant.relativity.greenend.org.uk>
Date: Sat, 27 Sep 2008 15:26:20 +0000 (+0100)
Subject: parallel thing compiles now
X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ian/git?p=moebius2.git;a=commitdiff_plain;h=21a2bd20afe0f4673494f4841a5675ca68cbd2d7

parallel thing compiles now
---

diff --git a/Makefile b/Makefile
index a0b643a..9443d70 100644
--- a/Makefile
+++ b/Makefile
@@ -80,8 +80,9 @@ output-%:	output+%.o mgraph+%.o common.o
 interpolate-%:	interpolate+%.o mgraph+%.o common.o
 		$(CC) $(CFLAGS) -o $@ $^ $(LIBGSL)
 
-minimise-%:	energy+%.o graph+%.o mgraph+%.o minimise+%.o half+%.o common.o
-		$(CXX) $(CXXFLAGS) -o $@ $^ $(LIBGSL)
+minimise-%:	energy+%.o graph+%.o mgraph+%.o minimise+%.o \
+			half+%.o parallel.o common.o
+		$(CXX) $(CXXFLAGS) -o $@ $^ $(LIBGSL) -lpthread
 
 # this ridiculous repetition is due to make being too lame
 
diff --git a/energy.c b/energy.c
index 80ac7e8..5aa1c60 100644
--- a/energy.c
+++ b/energy.c
@@ -5,51 +5,97 @@
 #include "common.h"
 #include "minimise.h"
 #include "mgraph.h"
+#include "parallel.h"
 
 double vertex_areas[N], vertex_mean_edge_lengths[N], edge_lengths[N][V6];
 
 static double best_energy= DBL_MAX;
 
 static void addcost(double *energy, double tweight, double tcost, int pr);
-#define COST(weight, compute) addcost(&energy, (weight), (compute), printing)
+
+/*---------- main energy computation, weights, etc. ----------*/
+
+typedef double CostComputation(const Vertices vertices, int section);
+
+typedef struct {
+  double weight;
+  CostComputation *fn;
+} CostContribution;
+
+static const CostContribution costs[]= {
+#define COST(weight, compute) { (weight),(compute) },
+
+#if XBITS==3
+#define STOP_EPSILON 1e-6;
+    COST(  3e2,   line_bending_cost)
+    COST(  1e3,   edge_length_variation_cost)
+    COST( 0.2e3,  rim_proximity_cost)
+    COST(  1e8,   noncircular_rim_cost)
+#endif
+
+#if XBITS==4
+#define STOP_EPSILON 1e-5;
+    COST(  3e2,   line_bending_cost)
+    COST(  3e3,   edge_length_variation_cost)
+    COST( 3.8e1,  rim_proximity_cost) // 5e1 is too much
+					 // 2.5e1 is too little
+    // 0.2e1 grows compared to previous ?
+    // 0.6e0 shrinks compared to previous ?
+    COST(  1e12,   noncircular_rim_cost)
+#endif
+};
+
+#define NCOSTS ((sizeof(costs)/sizeof(costs[0])))
 
 void energy_init(void) {
+  stop_epsilon= STOP_EPSILON;
+}
+
+void compute_energy_separately(const struct Vertices *vs,
+			 int section, void *energies_v, void *totals_v) {
+  double *energies= energies_v;
+  int ci;
+    
+  compute_edge_lengths(vs->a, section);
+  compute_vertex_areas(vs->a, section);
+
+  for (ci=0; ci<NCOSTS; ci++)
+    energies[ci]= costs[ci].fn(vs->a, section);
 }
 
-/*---------- main energy computation and subroutines ----------*/
+/*---------- energy computation machinery ----------*/
+
+void compute_energy_combine(const struct Vertices *vertices,
+			 int section, void *energies_v, void *totals_v) {
+  int ci;
+  
+  double *energies= energies_v;
+  double *totals= totals_v;
+
+  for (ci=0; ci<NCOSTS; ci++)
+    totals[ci] += energies[ci];
+}
 
 double compute_energy(const struct Vertices *vs) {
   static int bests_unprinted;
-  
-  double energy;
-  int printing;
 
-  compute_edge_lengths(vs->a);
-  compute_vertex_areas(vs->a);
-  energy= 0;
+  double totals[NCOSTS], energy;
+  int ci, printing;
 
   printing= printing_check(pr_cost,0);
 
   if (printing) printf("%15lld c>e |", evaluations);
 
-  if (XBITS==3) {
-    COST(  3e2,   line_bending_cost(vs->a));
-    COST(  1e3,   edge_length_variation_cost(vs->a));
-    COST( 0.2e3,  rim_proximity_cost(vs->a));
-    COST(  1e8,   noncircular_rim_cost(vs->a));
-    stop_epsilon= 1e-6;
-  } else if (XBITS==4) {
-    COST(  3e2,   line_bending_cost(vs->a));
-    COST(  3e3,   edge_length_variation_cost(vs->a));
-    COST( 3.8e1,  rim_proximity_cost(vs->a)); // 5e1 is too much
-						 // 2.5e1 is too little
-    // 0.2e1 grows compared to previous ?
-    // 0.6e0 shrinks compared to previous ?
-    COST(  1e12,   noncircular_rim_cost(vs->a));
-    stop_epsilon= 1e-5;
-  } else {
-    abort();
-  }
+  inparallel(vs,
+	     compute_energy_separately,
+	     compute_energy_combine,
+	     sizeof(totals) /* really, size of energies */,
+	     totals);
+
+  energy= 0;
+
+  for (ci=0; ci<NCOSTS; ci++)
+    addcost(&energy, costs[ci].weight, totals[ci], printing);
 
   if (printing) printf("| total %# e |", energy);
 
@@ -89,18 +135,18 @@ static void addcost(double *energy, double tweight, double tcost, int pr) {
 
 /*---------- Precomputations ----------*/
 
-void compute_edge_lengths(const Vertices vertices) {
+void compute_edge_lengths(const Vertices vertices, int section) {
   int v1,e,v2;
 
-  FOR_EDGE(v1,e,v2)
+  FOR_EDGE(v1,e,v2,OUTER)
     edge_lengths[v1][e]= hypotD(vertices[v1],vertices[v2]);
 }
 
-void compute_vertex_areas(const Vertices vertices) {
+void compute_vertex_areas(const Vertices vertices, int section) {
   int v0,v1,v2, e1,e2;
 //  int k;
 
-  FOR_VERTEX(v0) {
+  FOR_VERTEX(v0,OUTER) {
     double total= 0.0, edges_total=0;
     int count= 0;
 
@@ -162,7 +208,7 @@ void compute_vertex_areas(const Vertices vertices) {
    *          Q,e
    */
 
-double line_bending_cost(const Vertices vertices) {
+double line_bending_cost(const Vertices vertices, int section) {
   static const double axb_epsilon= 1e-6;
   static const double exponent_r= 3;
 
@@ -170,7 +216,7 @@ double line_bending_cost(const Vertices vertices) {
   double  a[D3], b[D3], axb[D3];
   double total_cost= 0;
 
-  FOR_EDGE(qi,e,ri) {
+  FOR_EDGE(qi,e,ri, OUTER) {
     pi= EDGE_END2(qi,(e+3)%V6); if (pi<0) continue;
 
     K a[k]= -vertices[pi][k] + vertices[qi][k];
@@ -200,11 +246,11 @@ double line_bending_cost(const Vertices vertices) {
    *           Q,e
    */
 
-double edge_length_variation_cost(const Vertices vertices) {
+double edge_length_variation_cost(const Vertices vertices, int section) {
   double diff, cost= 0, exponent_r= 2;
   int q, e,r, eback;
 
-  FOR_EDGE(q,e,r) {
+  FOR_EDGE(q,e,r, OUTER) {
     eback= edge_reverse(q,e);
     diff= edge_lengths[q][e] - edge_lengths[q][eback];
     cost += pow(diff,exponent_r);
@@ -225,11 +271,11 @@ static void find_nearest_oncircle(double oncircle[D3], const double p[D3]) {
   oncircle[1] *= mult;
 }
 
-double rim_proximity_cost(const Vertices vertices) {
+double rim_proximity_cost(const Vertices vertices, int section) {
   double oncircle[3], cost=0;
   int v;
 
-  FOR_VERTEX(v) {
+  FOR_VERTEX(v, OUTER) {
     int y= v >> YSHIFT;
     int nominal_edge_distance= y <= Y/2 ? y : Y-1-y;
     if (nominal_edge_distance==0) continue;
@@ -246,12 +292,12 @@ double rim_proximity_cost(const Vertices vertices) {
 
 /*---------- noncircular rim cost ----------*/
 
-double noncircular_rim_cost(const Vertices vertices) {
+double noncircular_rim_cost(const Vertices vertices, int section) {
   int vy,vx,v;
   double cost= 0.0;
   double oncircle[3];
 
-  FOR_RIM_VERTEX(vy,vx,v) {
+  FOR_RIM_VERTEX(vy,vx,v, OUTER) {
     find_nearest_oncircle(oncircle, vertices[v]);
 
     double d2= hypotD2(vertices[v], oncircle);
diff --git a/graph.c b/graph.c
index f0ce19d..a81de42 100644
--- a/graph.c
+++ b/graph.c
@@ -4,6 +4,7 @@
 
 #include "mgraph.h"
 #include "minimise.h"
+#include "parallel.h"
 
 static int sqdistances[N][N];
 
@@ -14,7 +15,7 @@ static void breadth_first_search(int start, int sqdistances_r[N]) {
   int v,e, current, future, dfuture;
 
   buf_push= buf_pop= buffer;
-  FOR_VERTEX(v) d[v]= -1;
+  FOR_VERTEX(v,INNER) d[v]= -1;
 
   d[start]= 0;
   *buf_push++= start;
@@ -31,7 +32,7 @@ static void breadth_first_search(int start, int sqdistances_r[N]) {
   assert(buf_pop==buf_push);
   assert(buf_push <= buffer+sizeof(buffer)/sizeof(buffer[0]));
 
-  FOR_VERTEX(v) {
+  FOR_VERTEX(v,INNER) {
     assert(d[v] >= 0);
     sqdistances_r[v]= d[v] * d[v];
   }
@@ -40,10 +41,10 @@ static void breadth_first_search(int start, int sqdistances_r[N]) {
 //    
 }
 
-void graph_layout_prepare() {
+void graph_layout_prepare(void) {
   int v1;
   
-  FOR_VERTEX(v1)
+  FOR_VERTEX(v1,INNER)
     breadth_first_search(v1, sqdistances[v1]);
 
   alpha= 2;
@@ -53,7 +54,7 @@ void graph_layout_prepare() {
 }
 
 
-double graph_layout_cost(const Vertices v) {
+double graph_layout_cost(const Vertices v, int section) {
   /* For each (vi,vj) computes shortest path s_ij = |vi..vj|
    * along edges, and actual distance d_ij = |vi-vj|.
    *
@@ -76,7 +77,7 @@ double graph_layout_cost(const Vertices v) {
   int v1,v2,e, nedges=0;
   double totaledgelength=0, meanedgelength, meanedgelength2;
 
-  FOR_EDGE(v1,e,v2) {
+  FOR_EDGE(v1,e,v2,INNER) {
     totaledgelength += hypotD(v[v1], v[v2]);
     nedges++;
   }
@@ -85,8 +86,8 @@ double graph_layout_cost(const Vertices v) {
   meanedgelength2= meanedgelength * meanedgelength;
 //  printf("mean=%g mean^2=%g\n", meanedgelength, meanedgelength2);
     
-  FOR_VERTEX(v1) {
-    FOR_VERTEX(v2) {
+  FOR_VERTEX(v1,OUTER) {
+    FOR_VERTEX(v2,INNER) {
       if (v1 == v2) continue;
 
       double d2= hypotD2(v[v1],v[v2]);
diff --git a/half.c b/half.c
index 5ac69f8..9051c73 100644
--- a/half.c
+++ b/half.c
@@ -57,7 +57,7 @@ void pmap_dimensions(const struct Vertices *vs) {
 
   bad= 0;
 
-  FOR_VERTEX(vc)
+  FOR_VERTEX(vc,INNER)
     FOR_COORD(kc) {
       count[vc][kc]= 0;
       ixfrom[vc][kc]= -1;
@@ -96,7 +96,7 @@ void pmap_dimensions(const struct Vertices *vs) {
 		    count[v][k]++;
 		    );
 
-  FOR_VERTEX(vc)
+  FOR_VERTEX(vc,INNER)
     FOR_COORD(kc) {
       if (count[vc][kc]==1) continue;
       printf("BAD %03x %d count=%d ixfrom=%d\n",
diff --git a/mgraph.h b/mgraph.h
index 06c53e2..7f2ab43 100644
--- a/mgraph.h
+++ b/mgraph.h
@@ -89,8 +89,11 @@
 #define V6 6
 #define V3 3
 
-#define FOR_VERTEX(v) \
-  for ((v)=0; (v)<N; (v)++)
+#define INNER(v,zero,n, precomp)  \
+  for ((v)=(zero); precomp, (v)<(n); (v)++)
+
+#define FOR_VERTEX(v,loop) \
+  loop ((v), 0, N, NOTHING)
 
 #define FOR_VPEDGE(e) \
   for ((e)=0; (e)<V6; (e)++)
@@ -117,13 +120,13 @@ int edge_reverse(int v1, int e);
 #define FOR_VEDGE(v1,e,v2)			\
   FOR_VEDGE_X(v1,e,v2,NOTHING,NOTHING)
 
-#define FOR_EDGE(v1,e,v2)			\
-  FOR_VERTEX((v1))				\
+#define FOR_EDGE(v1,e,v2, loop)			\
+  FOR_VERTEX((v1), loop)			\
     FOR_VEDGE((v1),(e),(v2))
 
-#define FOR_RIM_VERTEX(vy,vx,v)					\
-  for ((vy)=0; (vy)<Y; (vy)+=Y-1)				\
-    for ((vx)=0; (v)= (vy)<<YSHIFT | (vx), (vx)<X; (vx)++)
+#define FOR_RIM_VERTEX(vy,vx,v, loop)		\
+  for ((vy)=0; (vy)<Y; (vy)+=Y-1)		\
+    loop ((vx), 0, X, (v)= (vy)<<YSHIFT | (vx))
 
 int vertices_span_join_p(int v0, int v1);
 
diff --git a/minimise.h b/minimise.h
index a076125..097b8ac 100644
--- a/minimise.h
+++ b/minimise.h
@@ -10,17 +10,18 @@
 double compute_energy(const struct Vertices *vs);
 void energy_init(void);
 
-double graph_layout_cost(const Vertices v);
+double graph_layout_cost(const Vertices v, int section);
 void graph_layout_prepare();
 
-void compute_vertex_areas(const Vertices vertices);
-void compute_edge_lengths(const Vertices vertices);
+void compute_vertex_areas(const Vertices vertices, int section);
+void compute_edge_lengths(const Vertices vertices, int section);
+
 extern double vertex_areas[N], vertex_mean_edge_lengths[N], edge_lengths[N][V6];
 
-double line_bending_cost(const Vertices vertices);
-double noncircular_rim_cost(const Vertices vertices);
-double edge_length_variation_cost(const Vertices vertices);
-double rim_proximity_cost(const Vertices vertices);
+double line_bending_cost(const Vertices vertices, int section);
+double noncircular_rim_cost(const Vertices vertices, int section);
+double edge_length_variation_cost(const Vertices vertices, int section);
+double rim_proximity_cost(const Vertices vertices, int section);
 
 extern const char *input_file, *best_file;
 extern char *best_file_tmp;
diff --git a/output.c b/output.c
index 262991a..3367e43 100644
--- a/output.c
+++ b/output.c
@@ -197,7 +197,7 @@ static double outvertex_coord_check(double value) {
 static void compute_outvertices(void) {
   int v0,k,side,ab,x,y;
 
-  FOR_VERTEX(v0) {
+  FOR_VERTEX(v0, INNER) {
     for (ab=0; ab<2; ab++) {
       int v1= EDGE_END2(v0, ab?5:0);
       int v2= EDGE_END2(v0, ab?0:1);
@@ -211,7 +211,7 @@ static void compute_outvertices(void) {
       K Ok(ovAB[v0][ab][1], centroid[k] - normal[k]);
     }
   }
-  FOR_VERTEX(v0) {
+  FOR_VERTEX(v0, INNER) {
     int vw= EDGE_END2(v0,3);
     int vnw= EDGE_END2(v0,2);
     int vsw= EDGE_END2(v0,4);
@@ -234,7 +234,7 @@ static void compute_outvertices(void) {
       K Ok(ovC[v0][side], in[v0][k] + adjust[k]);
     }
   }
-  FOR_RIM_VERTEX(y,x,v0) {
+  FOR_RIM_VERTEX(y,x,v0, INNER) {
     double rim[D3], inner[D3], radius_cos[D3], radius_sin[D3];
     int vback, vfwd, around;
 
@@ -275,7 +275,7 @@ static void compute_outvertices(void) {
 	   sin(angle) * radius_sin[k]);
     }
   }
-  FOR_RIM_VERTEX(y,x,v0) {
+  FOR_RIM_VERTEX(y,x,v0, INNER) {
     int vfwd= EDGE_END2(v0,0);
     assert(vfwd >= 0);
     int aroung;
@@ -315,7 +315,7 @@ static OutVertex *invertex2outvertexcd(v0,side) {
 static void outfacets(void) {
   int v0,e,side,aroung;
   
-  FOR_VERTEX(v0) {
+  FOR_VERTEX(v0, INNER) {
     OutVertex *defs=0, *defs1=0;
     int rimy=-1;
     int_map *defs1aroundmap= 0;
diff --git a/parallel.c b/parallel.c
new file mode 100644
index 0000000..3c94607
--- /dev/null
+++ b/parallel.c
@@ -0,0 +1,61 @@
+/*
+ * Parallel processing
+ */
+
+#include <pthread.h>
+
+#include "mgraph.h"
+#include "parallel.h"
+
+typedef struct {
+  const struct Vertices *vertices;
+  Computation *separately;
+  void *gendata;
+} ForAllThreads;
+
+typedef struct {
+  ForAllThreads *allthreads;
+  int section;
+  void *secdata;
+  pthread_t thread;
+} PerThread;
+
+static void *routine(void *thread_v) {
+  PerThread *t= thread_v;
+  ForAllThreads *a= t->allthreads;
+
+  a->separately(a->vertices, t->section, t->secdata, a->gendata);
+
+  return 0;
+}
+
+void inparallel(const struct Vertices *vertices,
+		Computation *separately,
+		Computation *combine,
+		size_t secdatasz, void *gendata) {
+  typedef struct { unsigned char secdata[secdatasz]; } SecData;
+
+  ForAllThreads allthreads;
+  SecData secdatas[nsections];
+  PerThread threads[nsections];
+
+  int s, r;
+
+  allthreads.vertices= vertices;
+  allthreads.separately= separately;
+  allthreads.gendata= gendata;
+
+  for (s=0; s<nsections; s++) {
+    threads[s].allthreads= &allthreads;
+    threads[s].section= s;
+    threads[s].secdata= secdatas[s].secdata;
+    r= pthread_create(&threads[s].thread,0,routine,&threads[s]);
+    if (r) diee("pthread_create");
+  }
+
+  for (s=0; s<nsections; s++) {
+    r= pthread_join(threads[s].thread, 0);
+    if (r) diee("pthread_join");
+    combine(vertices, s, threads[s].secdata, gendata);
+  }
+}
diff --git a/parallel.h b/parallel.h
new file mode 100644
index 0000000..3d50822
--- /dev/null
+++ b/parallel.h
@@ -0,0 +1,34 @@
+/*
+ * Parallel processing
+ */
+
+#ifndef PARALLEL_H
+#define PARALLEL_H
+
+#ifdef NPROCESSORS
+# define NSECTIONS NPROCESSORS
+#else
+# define NSECTIONS 1
+#endif
+
+/* used anamorphically: section, nsections */
+
+#define OUTER_PERSECTION_BASE(zero,n, sect) \
+  ((zero) + sect * (((n)-(zero) + NSECTIONS-1) / NSECTIONS))
+#define OUTER(v,zero,n, precomp)					  \
+  for ((v)= OUTER_PERSECTION_BASE((zero),(n), section);			  \
+       precomp,								  \
+       (v) < OUTER_PERSECTION_BASE((zero),(n), section + 1) && (v) < (n); \
+       (v)++)
+
+#define nsections NSECTIONS
+
+typedef void Computation(const struct Vertices *vertices,
+			 int section, void *secdata, void *gendata);
+
+void inparallel(const struct Vertices *vertices,
+		Computation *separately,
+		Computation *combine,
+		size_t secdatasz, void *gendata);
+
+#endif /*PARALLEL_H*/
diff --git a/view.c b/view.c
index 32ae7d7..4f052c1 100644
--- a/view.c
+++ b/view.c
@@ -45,7 +45,7 @@ static void transform_coordinates(void) {
 
   int v, k;
   
-  FOR_VERTEX(v) {
+  FOR_VERTEX(v,INNER) {
     input_gsl.data= &conformation[v][0];
     GA( gsl_blas_dgemv(CblasNoTrans, 1.0,&transform_gsl, &input_gsl,
 		       0.0, &result_gsl) );
@@ -77,7 +77,7 @@ static void generate_display_list(void) {
   int vb, ve[V6], e;
 
   ntris= 0;
-  FOR_VERTEX(vb) {
+  FOR_VERTEX(vb,INNER) {
     /* We use the two triangles in the parallelogram vb, vb+e5, vb+e0, vb+e1.
      * We go round each triangle clockwise (although our surface is non-
      * orientable so it shouldn't matter).  Picking the parallelogram
@@ -94,7 +94,7 @@ static void generate_display_list(void) {
 
   if (!vertex_in_triangles_checked) {
     int v, expd;
-    FOR_VERTEX(v) {
+    FOR_VERTEX(v,INNER) {
       expd= RIM_VERTEX_P(v) ? 3 : 6;
       if (vertex_in_triangles[v] != expd) {
 	fprintf(stderr,"vertex %02x used for %d triangles, expected %d\n",
@@ -656,7 +656,7 @@ static void check_input(void) {
 
 static void topocheck(void) {
   int v1,e,v2,eprime,v1prime, count;
-  FOR_EDGE(v1,e,v2) {
+  FOR_EDGE(v1,e,v2, INNER) {
     count= 0;
     FOR_VEDGE(v2,eprime,v1prime)
       if (v1prime==v1) count++;