increase trims to prevent 1 in 100000 run crash

GPU.md

@@ -56,7 +56,7 @@ Here's a typical solver run:
 
     $ ./cuda30 -r 2
     GeForce GTX 1080 Ti with 10GB @ 352 bits x 5505MHz
-    Looking for 42-cycle on cuckoo30("",0-1) with 50% edges, 128*128 buckets, 240 trims, and 128 thread blocks.
+    Looking for 42-cycle on cuckoo30("",0-1) with 50% edges, 128*128 buckets, 256 trims, and 128 thread blocks.
     Using 2680MB bucket memory and 21MB memory per thread block (5392MB total)
     nonce 0 k0 k1 k2 k3 a34c6a2bdaa03a14 d736650ae53eee9e 9a22f05e3bffed5e b8d55478fa3a606d
        4-cycle found
@@ -82,7 +82,7 @@ bit key for the siphash function which generates the half-billion graph edges.
 This key is shown after each nonce as four 64-bit hexadecimal numbers. The GPU is
 responsible for generating all edges and then trimming the majority of
 them away for clearly not being part of any cycle. After a default number of
-240 trimming rounds, only about 1 in every 13450 edges survives, and the
+256 trimming rounds, only about 1 in every 13450 edges survives, and the
 remaining 37000 or so edges are sent back to the CPU for cycle finding, using a an
 algorithm inspired by Cuckoo Hashing (which is where the name derives from).
 
@@ -92,7 +92,7 @@ To see a synopsis of all possible options, run:
     SYNOPSIS
       cuda30 [-b blocks] [-d device] [-h hexheader] [-k rounds [-c count]] [-m trims] [-n nonce] [-r range] [-U blocks] [-u threads] [-V threads] [-v threads] [-T threads] [-t threads] [-X threads] [-x threads] [-Y threads] [-y threads] [-Z threads] [-z threads]
     DEFAULTS
-      cuda30 -b 128 -d 0 -h "" -k 0 -c 1 -m 240 -n 0 -r 1 -U 128 -u 8 -V 32 -v 128 -T 32 -t 128 -X 32 -x 64 -Y 32 -y 128 -Z 32 -z 8
+      cuda30 -b 128 -d 0 -h "" -k 0 -c 1 -m 256 -n 0 -r 1 -U 128 -u 8 -V 32 -v 128 -T 32 -t 128 -X 32 -x 64 -Y 32 -y 128 -Z 32 -z 8
 
 
 Most of these are for shaping the GPU's thread parallellism in the various edge generation and trimming rounds. 
@@ -102,7 +102,7 @@ Here's a run that uncovers a solution:
 
     $ ./cuda30 -n 60 -r 4
     GeForce GTX 1080 Ti with 10GB @ 352 bits x 5505MHz
-    Looking for 42-cycle on cuckoo30("",60-63) with 50% edges, 128*128 buckets, 240 trims, and 128 thread blocks.
+    Looking for 42-cycle on cuckoo30("",60-63) with 50% edges, 128*128 buckets, 256 trims, and 128 thread blocks.
     Using 2680MB bucket memory and 21MB memory per thread block (5392MB total)
     nonce 60 k0 k1 k2 k3 275f9313c78adcec c3dc47d972920e25 41f8c5d51abbf1e7 74da5cc5b52b2a0b
        6-cycle found

GPU_tuning.md

@@ -6,7 +6,7 @@ Recall the solver options:
     SYNOPSIS
       cuda30 [-b blocks] [-d device] [-h hexheader] [-k rounds [-c count]] [-m trims] [-n nonce] [-r range] [-U blocks] [-u threads] [-V threads] [-v threads] [-T threads] [-t threads] [-X threads] [-x threads] [-Y threads] [-y threads] [-Z threads] [-z threads]
     DEFAULTS
-      cuda30 -b 128 -d 0 -h "" -k 0 -c 1 -m 240 -n 0 -r 1 -U 128 -u 8 -V 32 -v 128 -T 32 -t 128 -X 32 -x 64 -Y 32 -y 128 -Z 32 -z 8
+      cuda30 -b 128 -d 0 -h "" -k 0 -c 1 -m 256 -n 0 -r 1 -U 128 -u 8 -V 32 -v 128 -T 32 -t 128 -X 32 -x 64 -Y 32 -y 128 -Z 32 -z 8
 
 Let's look at each of these in turn.
 
@@ -40,12 +40,12 @@ For example,
 
     $ ./cuda30 -h "DEADBEEF" | head -2
     GeForce GTX 1080 Ti with 10GB @ 352 bits x 5505MHz
-    Looking for 42-cycle on cuckoo30("ޭ??",0) with 50% edges, 128*128 buckets, 240 trims, and 128 thread blocks.
+    Looking for 42-cycle on cuckoo30("ޭ??",0) with 50% edges, 128*128 buckets, 256 trims, and 128 thread blocks.
 
     $ ./cuda30 -h "444541440A42454546" | head -3
     GeForce GTX 1080 Ti with 10GB @ 352 bits x 5505MHz
     Looking for 42-cycle on cuckoo30("DEAD
-    BEEF",0) with 50% edges, 128*128 buckets, 240 trims, and 128 thread blocks.
+    BEEF",0) with 50% edges, 128*128 buckets, 256 trims, and 128 thread blocks.
 
 -k rounds [-c counts]
 ------------
@@ -70,14 +70,14 @@ For example:
     round 13 size 542 completed in 2 ms
     round 14 size 470 completed in 3 ms
     round 15 size 403 completed in 2 ms
-    rounds 12 through 237 completed in 58 ms
-    trimrename3 round 238 size 2 completed in 1 ms
-    trimrename3 round 239 size 2 completed in 1 ms
+    rounds 12 through 253 completed in 58 ms
+    trimrename3 round 254 size 2 completed in 1 ms
+    trimrename3 round 255 size 2 completed in 1 ms
        4-cycle found
      282-cycle found
     1006-cycle found
      390-cycle found
-    findcycles completed on 33234 edges
+    findcycles completed on 29180 edges
     Time: 1010 ms
     0 total solutions
 
@@ -103,14 +103,14 @@ count all 128\*128 buckets is rather slow, as witnessed by the 3x longer Time be
     round 13 size 9301758 completed in 2 ms
     round 14 size 8167240 completed in 3 ms
     round 15 size 7230054 completed in 2 ms
-    rounds 12 through 237 completed in 709 ms
-    trimrename3 round 238 size 33514 completed in 1 ms
-    trimrename3 round 239 size 33514 completed in 1 ms
+    rounds 12 through 253 completed in 709 ms
+    trimrename3 round 254 size 33514 completed in 1 ms
+    trimrename3 round 255 size 33514 completed in 1 ms
        4-cycle found
      282-cycle found
     1006-cycle found
      390-cycle found
-    findcycles completed on 33234 edges
+    findcycles completed on 29180 edges
     Time: 3356 ms
     0 total solutions
 
@@ -128,7 +128,7 @@ back to the host.
 
 -m trims
 ------------
-The number of trimming rounds. Default 240. Can be increased arbitrarily. At some point, there will be
+The number of trimming rounds. Default 256. Can be increased arbitrarily. At some point, there will be
 no edges left to trim, as all remaining edges are already part of a cycle:
 
     $ ./cuda30 -m 1518 -k 2000 -c 128
@@ -149,8 +149,8 @@ no edges left to trim, as all remaining edges are already part of a cycle:
     round 1514 size 88 completed in 0 ms
     round 1515 size 88 completed in 0 ms
     rounds 12 through 1515 completed in 281145 ms
-    trimrename3 round 1998 size 88 completed in 0 ms
-    trimrename3 round 1999 size 88 completed in 0 ms
+    trimrename3 round 1516 size 88 completed in 0 ms
+    trimrename3 round 1517 size 88 completed in 0 ms
        4-cycle found
        2-cycle found
       16-cycle found

src/mean_miner.cu

@@ -223,7 +223,7 @@ struct trimparams {
   u16 reportrounds;
 
   trimparams() {
-    ntrims              = 240;
+    ntrims              = 256;
     nblocks             = 128;
     genUblocks          = 128;
     genUtpb             =   8;