ecmult_gen_impl: eliminate scratch memory used when generating context

Author: apoelstra

src/bench_ecmult.c

@@ -172,7 +172,7 @@ int main(int argc, char **argv) {
             secp256k1_scalar_add(&data.seckeys[i], &data.seckeys[i - 1], &data.seckeys[i - 1]);
         }
     }
-    secp256k1_ge_set_all_gej_var(data.pubkeys, pubkeys_gej, POINTS, &data.ctx->error_callback);
+    secp256k1_ge_set_all_gej_var(data.pubkeys, pubkeys_gej, POINTS);
     free(pubkeys_gej);
 
     for (i = 1; i <= 8; ++i) {

src/ecmult_gen_impl.h

@@ -77,7 +77,7 @@ static void secp256k1_ecmult_gen_context_build(secp256k1_ecmult_gen_context *ctx
                 secp256k1_gej_add_var(&numsbase, &numsbase, &nums_gej, NULL);
             }
         }
-        secp256k1_ge_set_all_gej_var(prec, precj, 1024, cb);
+        secp256k1_ge_set_all_gej_var(prec, precj, 1024);
     }
     for (j = 0; j < 64; j++) {
         for (i = 0; i < 16; i++) {

src/group.h

@@ -65,7 +65,7 @@ static void secp256k1_ge_neg(secp256k1_ge *r, const secp256k1_ge *a);
 static void secp256k1_ge_set_gej(secp256k1_ge *r, secp256k1_gej *a);
 
 /** Set a batch of group elements equal to the inputs given in jacobian coordinates */
-static void secp256k1_ge_set_all_gej_var(secp256k1_ge *r, const secp256k1_gej *a, size_t len, const secp256k1_callback *cb);
+static void secp256k1_ge_set_all_gej_var(secp256k1_ge *r, const secp256k1_gej *a, size_t len);
 
 /** Set a batch of group elements equal to the inputs given in jacobian
  *  coordinates (with known z-ratios). zr must contain the known z-ratios such

src/group_impl.h

@@ -126,30 +126,45 @@ static void secp256k1_ge_set_gej_var(secp256k1_ge *r, secp256k1_gej *a) {
     r->y = a->y;
 }
 
-static void secp256k1_ge_set_all_gej_var(secp256k1_ge *r, const secp256k1_gej *a, size_t len, const secp256k1_callback *cb) {
-    secp256k1_fe *az;
-    secp256k1_fe *azi;
+static void secp256k1_ge_set_all_gej_var(secp256k1_ge *r, const secp256k1_gej *a, size_t len) {
+    secp256k1_fe u;
     size_t i;
-    size_t count = 0;
-    az = (secp256k1_fe *)checked_malloc(cb, sizeof(secp256k1_fe) * len);
+    size_t last_i = SIZE_MAX;
+
     for (i = 0; i < len; i++) {
         if (!a[i].infinity) {
-            az[count++] = a[i].z;
+            /* Use destination's x coordinates as scratch space */
+            if (last_i == SIZE_MAX) {
+                r[i].x = a[i].z;
+            } else {
+                secp256k1_fe_mul(&r[i].x, &r[last_i].x, &a[i].z);
+            }
+            last_i = i;
         }
     }
+    if (last_i == SIZE_MAX) {
+        return;
+    }
+    secp256k1_fe_inv_var(&u, &r[last_i].x);
 
-    azi = (secp256k1_fe *)checked_malloc(cb, sizeof(secp256k1_fe) * count);
-    secp256k1_fe_inv_all_var(azi, az, count);
-    free(az);
+    i = last_i;
+    while (i > 0) {
+        i--;
+        if (!a[i].infinity) {
+            secp256k1_fe_mul(&r[last_i].x, &r[i].x, &u);
+            secp256k1_fe_mul(&u, &u, &a[last_i].z);
+            last_i = i;
+        }
+    }
+    VERIFY_CHECK(!a[last_i].infinity);
+    r[last_i].x = u;
 
-    count = 0;
     for (i = 0; i < len; i++) {
         r[i].infinity = a[i].infinity;
         if (!a[i].infinity) {
-            secp256k1_ge_set_gej_zinv(&r[i], &a[i], &azi[count++]);
+            secp256k1_ge_set_gej_zinv(&r[i], &a[i], &r[i].x);
         }
     }
-    free(azi);
 }
 
 static void secp256k1_ge_set_table_gej_var(secp256k1_ge *r, const secp256k1_gej *a, const secp256k1_fe *zr, size_t len) {

src/tests.c

@@ -2104,7 +2104,7 @@ void test_ge(void) {
             }
         }
         secp256k1_ge_set_table_gej_var(ge_set_table, gej, zr, 4 * runs + 1);
-        secp256k1_ge_set_all_gej_var(ge_set_all, gej, 4 * runs + 1, &ctx->error_callback);
+        secp256k1_ge_set_all_gej_var(ge_set_all, gej, 4 * runs + 1);
         for (i = 0; i < 4 * runs + 1; i++) {
             secp256k1_fe s;
             random_fe_non_zero(&s);