diff --git a/ed25519/amd64-51-30k/fe25519.h b/ed25519/amd64-51-30k/fe25519.h
index dfad55e..8a16cd8 100644
--- a/ed25519/amd64-51-30k/fe25519.h
+++ b/ed25519/amd64-51-30k/fe25519.h
@@ -1,6 +1,8 @@
 #ifndef FE25519_H
 #define FE25519_H
 
+#include <stddef.h>
+
 #define fe25519              crypto_sign_ed25519_amd64_51_30k_batch_fe25519
 #define fe25519_freeze       crypto_sign_ed25519_amd64_51_30k_batch_fe25519_freeze
 #define fe25519_unpack       crypto_sign_ed25519_amd64_51_30k_batch_fe25519_unpack
@@ -60,6 +62,8 @@ void fe25519_nsquare(fe25519 *r, unsigned long long n);
 
 void fe25519_invert(fe25519 *r, const fe25519 *x);
 
+void fe25519_batchinvert(fe25519 *out[],fe25519 tmp[],fe25519 * const in[], size_t num);
+
 void fe25519_pow2523(fe25519 *r, const fe25519 *x);
 
 #endif
diff --git a/ed25519/amd64-51-30k/fe25519_batchinvert.c b/ed25519/amd64-51-30k/fe25519_batchinvert.c
new file mode 100644
index 0000000..b29387a
--- /dev/null
+++ b/ed25519/amd64-51-30k/fe25519_batchinvert.c
@@ -0,0 +1,26 @@
+#include "fe25519.h"
+
+// tmp MUST != out
+// in MAY == out
+void fe25519_batchinvert(fe25519 *out[],fe25519 tmp[],fe25519 * const in[], size_t num)
+{
+  fe25519 acc;
+  fe25519 tmpacc;
+  size_t i;
+
+  fe25519_setint(&acc,1);
+
+  for (i = 0;i < num;++i) {
+    tmp[i] = acc;
+    fe25519_mul(&acc,&acc,in[i]);
+  }
+
+  fe25519_invert(&acc,&acc);
+
+  i = num;
+  while (i--) {
+    fe25519_mul(&tmpacc,&acc,in[i]);
+    fe25519_mul(out[i],&acc,&tmp[i]);
+    acc = tmpacc;
+  }
+}
diff --git a/ed25519/amd64-51-30k/ge25519.h b/ed25519/amd64-51-30k/ge25519.h
index 02fd73a..5628090 100644
--- a/ed25519/amd64-51-30k/ge25519.h
+++ b/ed25519/amd64-51-30k/ge25519.h
@@ -73,6 +73,8 @@ typedef struct
   fe25519 t2d;
 } ge25519_pniels;
 
+typedef unsigned char bytes32[32];
+
 extern void ge25519_p1p1_to_p2(ge25519_p2 *r, const ge25519_p1p1 *p);
 extern void ge25519_p1p1_to_p3(ge25519_p3 *r, const ge25519_p1p1 *p);
 extern void ge25519_p1p1_to_pniels(ge25519_pniels *r, const ge25519_p1p1 *p);
@@ -90,6 +92,9 @@ extern int ge25519_unpackneg_vartime(ge25519 *r, const unsigned char p[32]);
 
 extern void ge25519_pack(unsigned char r[32], const ge25519 *p);
 
+extern void ge25519_batchpack_destructive_1(bytes32 out[], ge25519_p3 in[], fe25519 *inz[], fe25519 tmp[], size_t num);
+extern void ge25519_batchpack_destructive_finish(bytes32 out, ge25519_p3 *unf);
+
 extern int ge25519_isneutral_vartime(const ge25519 *p);
 
 extern void ge25519_add(ge25519 *r, const ge25519 *p, const ge25519 *q);
diff --git a/ed25519/amd64-51-30k/ge25519_batchpack.c b/ed25519/amd64-51-30k/ge25519_batchpack.c
new file mode 100644
index 0000000..d0d8f95
--- /dev/null
+++ b/ed25519/amd64-51-30k/ge25519_batchpack.c
@@ -0,0 +1,24 @@
+#include "fe25519.h"
+#include "ge25519.h"
+
+// assumes inz[] points to things in in[]
+// NOTE: leaves in unfinished state
+void ge25519_batchpack_destructive_1(bytes32 out[], ge25519_p3 in[], fe25519 *inz[], fe25519 tmp[], size_t num)
+{
+  fe25519 ty;
+
+  fe25519_batchinvert(inz, tmp, inz, num);
+
+  for (size_t i = 0; i < num; ++i) {
+    fe25519_mul(&ty, &in[i].y, &in[i].z);
+    fe25519_pack(out[i], &ty);
+  }
+}
+
+void ge25519_batchpack_destructive_finish(bytes32 out, ge25519_p3 *unf)
+{
+  fe25519 tx;
+  // z of unfinished is inverted
+  fe25519_mul(&tx, &unf->x, &unf->z);
+  out[31] ^= fe25519_getparity(&tx) << 7;
+}
diff --git a/main.c b/main.c
index b374930..cea5839 100644
--- a/main.c
+++ b/main.c
@@ -520,7 +520,7 @@ end:
 #ifdef BATCHKEYGEN
 
 #ifndef BATCHNUM
-#define BATCHNUM 256
+#define BATCHNUM 2048
 #endif
 
 static void *dobatchwork(void *task)
@@ -1045,6 +1045,7 @@ int main(int argc,char **argv)
 	VEC_ZERO(tstats);
 #endif
 
+#if 0
 	pthread_attr_t tattr,*tattrp = &tattr;
 	tret = pthread_attr_init(tattrp);
 	if (tret) {
@@ -1056,13 +1057,14 @@ int main(int argc,char **argv)
 		if (tret)
 			perror("pthread_attr_setstacksize");
 	}
+#endif
 
 	for (size_t i = 0;i < VEC_LENGTH(threads);++i) {
 		void *tp = 0;
 #ifdef STATISTICS
 		tp = &VEC_BUF(stats,i);
 #endif
-		tret = pthread_create(&VEC_BUF(threads,i),tattrp,
+		tret = pthread_create(&VEC_BUF(threads,i),0,
 #ifdef PASSPHRASE
 				deterministic ? dofastworkdeterministic :
 #endif
@@ -1076,11 +1078,13 @@ int main(int argc,char **argv)
 		}
 	}
 
+#if 0
 	if (tattrp) {
 		tret = pthread_attr_destroy(tattrp);
 		if (tret)
 			perror("pthread_attr_destroy");
 	}
+#endif
 
 #ifdef STATISTICS
 	struct timespec nowtime;