Context isn't freed in the ECDH benchmark

Summary:
 * free the ctx at the end of bench_ecdh

 * Pass num of iters to benchmarks as variable, and define envvar

 * Add running benchmarks regularly and under valgrind in travis

This is a backport of libsecp256k1 [[bitcoin-core/secp256k1#722 | PR722]]

Test Plan:
  ninja all check check-secp256k1 bench-secp256k1

Reviewers: #bitcoin_abc, Fabien

Reviewed By: #bitcoin_abc, Fabien

Subscribers: Fabien

Differential Revision: https://reviews.bitcoinabc.org/D6364

src/secp256k1/.travis.yml

@@ -41,6 +41,8 @@ env:
     - OPENSSL_TESTS=auto
     - MULTISET=no
     - CTIMETEST=yes
+    - BENCH=yes
+    - SECP256K1_BENCH_ITERS=2
   jobs:
     - SCALAR=32bit    RECOVERY=yes
     - SCALAR=32bit    FIELD=32bit       ECDH=yes  EXPERIMENTAL=yes  MULTISET=yes
@@ -54,10 +56,10 @@ env:
     - BIGNUM=no
     - BIGNUM=no       ENDOMORPHISM=yes  RECOVERY=yes  EXPERIMENTAL=yes  MULTISET=yes
     - BIGNUM=no       STATICPRECOMPUTATION=no
-    - AUTOTOOLS_TARGET=distcheck       CMAKE_TARGET=install    CTIMETEST=
+    - AUTOTOOLS_TARGET=distcheck       CMAKE_TARGET=install    CTIMETEST= BENCH=
     - AUTOTOOLS_EXTRA_FLAGS=CPPFLAGS=-DDETERMINISTIC CMAKE_EXTRA_FLAGS=-DCMAKE_C_FLAGS=-DDETERMINISTIC
     - AUTOTOOLS_EXTRA_FLAGS=CFLAGS=-O0               CMAKE_EXTRA_FLAGS=-DCMAKE_BUILD_TYPE=Debug
-    - AUTOTOOLS_TARGET=check-java      CMAKE_TARGET=check-secp256k1-java JNI=yes ECDH=yes EXPERIMENTAL=yes  CTIMETEST=
+    - AUTOTOOLS_TARGET=check-java      CMAKE_TARGET=check-secp256k1-java JNI=yes ECDH=yes EXPERIMENTAL=yes CTIMETEST= BENCH=
     - ECMULTGENPRECISION=2
     - ECMULTGENPRECISION=8
     - SCHNORR=no

src/secp256k1/src/bench.h

@@ -73,7 +73,7 @@ void print_number(const int64_t x) {
     printf("%s", &buffer[ptr]);
 }
 
-void run_benchmark(char *name, void (*benchmark)(void*), void (*setup)(void*), void (*teardown)(void*), void* data, int count, int iter) {
+void run_benchmark(char *name, void (*benchmark)(void*, int), void (*setup)(void*), void (*teardown)(void*, int), void* data, int count, int iter) {
     int i;
     int64_t min = INT64_MAX;
     int64_t sum = 0;
@@ -84,10 +84,10 @@ void run_benchmark(char *name, void (*benchmark)(void*), void (*setup)(void*), v
             setup(data);
         }
         begin = gettime_i64();
-        benchmark(data);
+        benchmark(data, iter);
         total = gettime_i64() - begin;
         if (teardown != NULL) {
-            teardown(data);
+            teardown(data, iter);
         }
         if (total < min) {
             min = total;
@@ -121,4 +121,13 @@ int have_flag(int argc, char** argv, char *flag) {
     return 0;
 }
 
+int get_iters(int default_iters) {
+    char* env = getenv("SECP256K1_BENCH_ITERS");
+    if (env) {
+        return strtol(env, NULL, 0);
+    } else {
+        return default_iters;
+    }
+}
+
 #endif /* SECP256K1_BENCH_H */

src/secp256k1/src/bench_ecdh.c

@@ -28,27 +28,32 @@ static void bench_ecdh_setup(void* arg) {
         0xa2, 0xba, 0xd1, 0x84, 0xf8, 0x83, 0xc6, 0x9f
     };
 
-    /* create a context with no capabilities */
-    data->ctx = secp256k1_context_create(SECP256K1_FLAGS_TYPE_CONTEXT);
     for (i = 0; i < 32; i++) {
         data->scalar[i] = i + 1;
     }
     CHECK(secp256k1_ec_pubkey_parse(data->ctx, &data->point, point, sizeof(point)) == 1);
 }
 
-static void bench_ecdh(void* arg) {
+static void bench_ecdh(void* arg, int iters) {
     int i;
     unsigned char res[32];
     bench_ecdh_data *data = (bench_ecdh_data*)arg;
 
-    for (i = 0; i < 20000; i++) {
+    for (i = 0; i < iters; i++) {
         CHECK(secp256k1_ecdh(data->ctx, res, &data->point, data->scalar, NULL, NULL) == 1);
     }
 }
 
 int main(void) {
     bench_ecdh_data data;
 
-    run_benchmark("ecdh", bench_ecdh, bench_ecdh_setup, NULL, &data, 10, 20000);
+    int iters = get_iters(20000);
+
+    /* create a context with no capabilities */
+    data.ctx = secp256k1_context_create(SECP256K1_FLAGS_TYPE_CONTEXT);
+
+    run_benchmark("ecdh", bench_ecdh, bench_ecdh_setup, NULL, &data, 10, iters);
+
+    secp256k1_context_destroy(data.ctx);
     return 0;
 }

src/secp256k1/src/bench_ecmult.c

@@ -18,7 +18,6 @@
 #include "secp256k1.c"
 
 #define POINTS 32768
-#define ITERS 10000
 
 typedef struct {
     /* Setup once in advance */
@@ -55,13 +54,13 @@ static int bench_callback(secp256k1_scalar* sc, secp256k1_ge* ge, size_t idx, vo
     return 1;
 }
 
-static void bench_ecmult(void* arg) {
+static void bench_ecmult(void* arg, int iters) {
     bench_data* data = (bench_data*)arg;
 
-    size_t count = data->count;
     int includes_g = data->includes_g;
-    size_t iters = 1 + ITERS / count;
-    size_t iter;
+    int iter;
+    int count = data->count;
+    iters = iters / data->count;
 
     for (iter = 0; iter < iters; ++iter) {
         data->ecmult_multi(&data->ctx->error_callback, &data->ctx->ecmult_ctx, data->scratch, &data->output[iter], data->includes_g ? &data->scalars[data->offset1] : NULL, bench_callback, arg, count - includes_g);
@@ -76,10 +75,10 @@ static void bench_ecmult_setup(void* arg) {
     data->offset2 = (data->count * 0x7f6f537b + 0x6a1a8f49) % POINTS;
 }
 
-static void bench_ecmult_teardown(void* arg) {
+static void bench_ecmult_teardown(void* arg, int iters) {
     bench_data* data = (bench_data*)arg;
-    size_t iters = 1 + ITERS / data->count;
-    size_t iter;
+    int iter;
+    iters = iters / data->count;
     /* Verify the results in teardown, to avoid doing comparisons while benchmarking. */
     for (iter = 0; iter < iters; ++iter) {
         secp256k1_gej tmp;
@@ -104,10 +103,10 @@ static void generate_scalar(uint32_t num, secp256k1_scalar* scalar) {
     CHECK(!overflow);
 }
 
-static void run_test(bench_data* data, size_t count, int includes_g) {
+static void run_test(bench_data* data, size_t count, int includes_g, int num_iters) {
     char str[32];
     static const secp256k1_scalar zero = SECP256K1_SCALAR_CONST(0, 0, 0, 0, 0, 0, 0, 0);
-    size_t iters = 1 + ITERS / count;
+    size_t iters = 1 + num_iters / count;
     size_t iter;
 
     data->count = count;
@@ -130,7 +129,7 @@ static void run_test(bench_data* data, size_t count, int includes_g) {
 
     /* Run the benchmark. */
     sprintf(str, includes_g ? "ecmult_%ig" : "ecmult_%i", (int)count);
-    run_benchmark(str, bench_ecmult, bench_ecmult_setup, bench_ecmult_teardown, data, 10, count * (1 + ITERS / count));
+    run_benchmark(str, bench_ecmult, bench_ecmult_setup, bench_ecmult_teardown, data, 10, count * iters);
 }
 
 int main(int argc, char **argv) {
@@ -139,6 +138,8 @@ int main(int argc, char **argv) {
     secp256k1_gej* pubkeys_gej;
     size_t scratch_size;
 
+    int iters = get_iters(10000);
+
     data.ctx = secp256k1_context_create(SECP256K1_CONTEXT_SIGN | SECP256K1_CONTEXT_VERIFY);
     scratch_size = secp256k1_strauss_scratch_size(POINTS) + STRAUSS_SCRATCH_OBJECTS*16;
     data.scratch = secp256k1_scratch_space_create(data.ctx, scratch_size);
@@ -167,8 +168,8 @@ int main(int argc, char **argv) {
     data.scalars = malloc(sizeof(secp256k1_scalar) * POINTS);
     data.seckeys = malloc(sizeof(secp256k1_scalar) * POINTS);
     data.pubkeys = malloc(sizeof(secp256k1_ge) * POINTS);
-    data.expected_output = malloc(sizeof(secp256k1_gej) * (ITERS + 1));
-    data.output = malloc(sizeof(secp256k1_gej) * (ITERS + 1));
+    data.expected_output = malloc(sizeof(secp256k1_gej) * (iters + 1));
+    data.output = malloc(sizeof(secp256k1_gej) * (iters + 1));
 
     /* Generate a set of scalars, and private/public keypairs. */
     pubkeys_gej = malloc(sizeof(secp256k1_gej) * POINTS);
@@ -185,14 +186,20 @@ int main(int argc, char **argv) {
     free(pubkeys_gej);
 
     for (i = 1; i <= 8; ++i) {
-        run_test(&data, i, 1);
+        run_test(&data, i, 1, iters);
     }
 
-    for (p = 0; p <= 11; ++p) {
-        for (i = 9; i <= 16; ++i) {
-            run_test(&data, i << p, 1);
+    /* This is disabled with low count of iterations because the loop runs 77 times even with iters=1
+    * and the higher it goes the longer the computation takes(more points)
+    * So we don't run this benchmark with low iterations to prevent slow down */
+     if (iters > 2) {
+        for (p = 0; p <= 11; ++p) {
+            for (i = 9; i <= 16; ++i) {
+                run_test(&data, i << p, 1, iters);
+            }
         }
     }
+
     if (data.scratch != NULL) {
         secp256k1_scratch_space_destroy(data.ctx, data.scratch);
     }