#define POINTS 32768
+void help(char **argv) {
+ printf("Benchmark EC multiplication algorithms\n");
+ printf("\n");
+ printf("Usage: %s <help|pippenger_wnaf|strauss_wnaf|simple>\n", argv[0]);
+ printf("The output shows the number of multiplied and summed points right after the\n");
+ printf("function name. The letter 'g' indicates that one of the points is the generator.\n");
+ printf("The benchmarks are divided by the number of points.\n");
+ printf("\n");
+ printf("default (ecmult_multi): picks pippenger_wnaf or strauss_wnaf depending on the\n");
+ printf(" batch size\n");
+ printf("pippenger_wnaf: for all batch sizes\n");
+ printf("strauss_wnaf: for all batch sizes\n");
+ printf("simple: multiply and sum each point individually\n");
+}
+
typedef struct {
/* Setup once in advance */
secp256k1_context* ctx;
secp256k1_scratch_space* scratch;
secp256k1_scalar* scalars;
secp256k1_ge* pubkeys;
+ secp256k1_gej* pubkeys_gej;
secp256k1_scalar* seckeys;
secp256k1_gej* expected_output;
secp256k1_ecmult_multi_func ecmult_multi;
data->offset2 = (x * 0x7f6f537b + 0x6a1a8f49) % POINTS;
}
+/* Check correctness of the benchmark by computing
+ * sum(outputs) ?= (sum(scalars_gen) + sum(seckeys)*sum(scalars))*G */
+static void bench_ecmult_teardown_helper(bench_data* data, size_t* seckey_offset, size_t* scalar_offset, size_t* scalar_gen_offset, int iters) {
+ int i;
+ secp256k1_gej sum_output, tmp;
+ secp256k1_scalar sum_scalars;
+
+ secp256k1_gej_set_infinity(&sum_output);
+ secp256k1_scalar_clear(&sum_scalars);
+ for (i = 0; i < iters; ++i) {
+ secp256k1_gej_add_var(&sum_output, &sum_output, &data->output[i], NULL);
+ if (scalar_gen_offset != NULL) {
+ secp256k1_scalar_add(&sum_scalars, &sum_scalars, &data->scalars[(*scalar_gen_offset+i) % POINTS]);
+ }
+ if (seckey_offset != NULL) {
+ secp256k1_scalar s = data->seckeys[(*seckey_offset+i) % POINTS];
+ secp256k1_scalar_mul(&s, &s, &data->scalars[(*scalar_offset+i) % POINTS]);
+ secp256k1_scalar_add(&sum_scalars, &sum_scalars, &s);
+ }
+ }
+ secp256k1_ecmult_gen(&data->ctx->ecmult_gen_ctx, &tmp, &sum_scalars);
+ secp256k1_gej_neg(&tmp, &tmp);
+ secp256k1_gej_add_var(&tmp, &tmp, &sum_output, NULL);
+ CHECK(secp256k1_gej_is_infinity(&tmp));
+}
+
+static void bench_ecmult_setup(void* arg) {
+ bench_data* data = (bench_data*)arg;
+ /* Re-randomize offset to ensure that we're using different scalars and
+ * group elements in each run. */
+ hash_into_offset(data, data->offset1);
+}
+
+static void bench_ecmult_gen(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ int i;
+
+ for (i = 0; i < iters; ++i) {
+ secp256k1_ecmult_gen(&data->ctx->ecmult_gen_ctx, &data->output[i], &data->scalars[(data->offset1+i) % POINTS]);
+ }
+}
+
+static void bench_ecmult_gen_teardown(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ bench_ecmult_teardown_helper(data, NULL, NULL, &data->offset1, iters);
+}
+
+static void bench_ecmult_const(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ int i;
+
+ for (i = 0; i < iters; ++i) {
+ secp256k1_ecmult_const(&data->output[i], &data->pubkeys[(data->offset1+i) % POINTS], &data->scalars[(data->offset2+i) % POINTS], 256);
+ }
+}
+
+static void bench_ecmult_const_teardown(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ bench_ecmult_teardown_helper(data, &data->offset1, &data->offset2, NULL, iters);
+}
+
+static void bench_ecmult_1(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ int i;
+
+ for (i = 0; i < iters; ++i) {
+ secp256k1_ecmult(&data->ctx->ecmult_ctx, &data->output[i], &data->pubkeys_gej[(data->offset1+i) % POINTS], &data->scalars[(data->offset2+i) % POINTS], NULL);
+ }
+}
+
+static void bench_ecmult_1_teardown(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ bench_ecmult_teardown_helper(data, &data->offset1, &data->offset2, NULL, iters);
+}
+
+static void bench_ecmult_1g(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ secp256k1_scalar zero;
+ int i;
+
+ secp256k1_scalar_set_int(&zero, 0);
+ for (i = 0; i < iters; ++i) {
+ secp256k1_ecmult(&data->ctx->ecmult_ctx, &data->output[i], NULL, &zero, &data->scalars[(data->offset1+i) % POINTS]);
+ }
+}
+
+static void bench_ecmult_1g_teardown(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ bench_ecmult_teardown_helper(data, NULL, NULL, &data->offset1, iters);
+}
+
+static void bench_ecmult_2g(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ int i;
+
+ for (i = 0; i < iters/2; ++i) {
+ secp256k1_ecmult(&data->ctx->ecmult_ctx, &data->output[i], &data->pubkeys_gej[(data->offset1+i) % POINTS], &data->scalars[(data->offset2+i) % POINTS], &data->scalars[(data->offset1+i) % POINTS]);
+ }
+}
+
+static void bench_ecmult_2g_teardown(void* arg, int iters) {
+ bench_data* data = (bench_data*)arg;
+ bench_ecmult_teardown_helper(data, &data->offset1, &data->offset2, &data->offset1, iters/2);
+}
+
+static void run_ecmult_bench(bench_data* data, int iters) {
+ char str[32];
+ sprintf(str, "ecmult_gen");
+ run_benchmark(str, bench_ecmult_gen, bench_ecmult_setup, bench_ecmult_gen_teardown, data, 10, iters);
+ sprintf(str, "ecmult_const");
+ run_benchmark(str, bench_ecmult_const, bench_ecmult_setup, bench_ecmult_const_teardown, data, 10, iters);
+ /* ecmult with non generator point */
+ sprintf(str, "ecmult 1");
+ run_benchmark(str, bench_ecmult_1, bench_ecmult_setup, bench_ecmult_1_teardown, data, 10, iters);
+ /* ecmult with generator point */
+ sprintf(str, "ecmult 1g");
+ run_benchmark(str, bench_ecmult_1g, bench_ecmult_setup, bench_ecmult_1g_teardown, data, 10, iters);
+ /* ecmult with generator and non-generator point. The reported time is per point. */
+ sprintf(str, "ecmult 2g");
+ run_benchmark(str, bench_ecmult_2g, bench_ecmult_setup, bench_ecmult_2g_teardown, data, 10, 2*iters);
+}
+
static int bench_ecmult_multi_callback(secp256k1_scalar* sc, secp256k1_ge* ge, size_t idx, void* arg) {
bench_data* data = (bench_data*)arg;
if (data->includes_g) ++idx;
int main(int argc, char **argv) {
bench_data data;
int i, p;
- 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);
data.ecmult_multi = secp256k1_ecmult_multi_var;
if (argc > 1) {
- if(have_flag(argc, argv, "pippenger_wnaf")) {
+ if(have_flag(argc, argv, "-h")
+ || have_flag(argc, argv, "--help")
+ || have_flag(argc, argv, "help")) {
+ help(argv);
+ return 1;
+ } else if(have_flag(argc, argv, "pippenger_wnaf")) {
printf("Using pippenger_wnaf:\n");
data.ecmult_multi = secp256k1_ecmult_pippenger_batch_single;
} else if(have_flag(argc, argv, "strauss_wnaf")) {
data.ecmult_multi = secp256k1_ecmult_strauss_batch_single;
} else if(have_flag(argc, argv, "simple")) {
printf("Using simple algorithm:\n");
- data.ecmult_multi = secp256k1_ecmult_multi_var;
- secp256k1_scratch_space_destroy(data.ctx, data.scratch);
- data.scratch = NULL;
} else {
- fprintf(stderr, "%s: unrecognized argument '%s'.\n", argv[0], argv[1]);
- fprintf(stderr, "Use 'pippenger_wnaf', 'strauss_wnaf', 'simple' or no argument to benchmark a combined algorithm.\n");
+ fprintf(stderr, "%s: unrecognized argument '%s'.\n\n", argv[0], argv[1]);
+ help(argv);
return 1;
}
}
+ data.ctx = secp256k1_context_create(SECP256K1_CONTEXT_SIGN | SECP256K1_CONTEXT_VERIFY);
+ scratch_size = secp256k1_strauss_scratch_size(POINTS) + STRAUSS_SCRATCH_OBJECTS*16;
+ if (!have_flag(argc, argv, "simple")) {
+ data.scratch = secp256k1_scratch_space_create(data.ctx, scratch_size);
+ } else {
+ data.scratch = NULL;
+ }
+
/* Allocate stuff */
data.scalars = malloc(sizeof(secp256k1_scalar) * POINTS);
data.seckeys = malloc(sizeof(secp256k1_scalar) * POINTS);
data.pubkeys = malloc(sizeof(secp256k1_ge) * POINTS);
+ data.pubkeys_gej = malloc(sizeof(secp256k1_gej) * POINTS);
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);
- secp256k1_gej_set_ge(&pubkeys_gej[0], &secp256k1_ge_const_g);
+ secp256k1_gej_set_ge(&data.pubkeys_gej[0], &secp256k1_ge_const_g);
secp256k1_scalar_set_int(&data.seckeys[0], 1);
for (i = 0; i < POINTS; ++i) {
generate_scalar(i, &data.scalars[i]);
if (i) {
- secp256k1_gej_double_var(&pubkeys_gej[i], &pubkeys_gej[i - 1], NULL);
+ secp256k1_gej_double_var(&data.pubkeys_gej[i], &data.pubkeys_gej[i - 1], NULL);
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);
- free(pubkeys_gej);
+ secp256k1_ge_set_all_gej_var(data.pubkeys, data.pubkeys_gej, POINTS);
+
+
+ /* Initialize offset1 and offset2 */
+ hash_into_offset(&data, 0);
+ run_ecmult_bench(&data, iters);
for (i = 1; i <= 8; ++i) {
run_ecmult_multi_bench(&data, i, 1, iters);
secp256k1_context_destroy(data.ctx);
free(data.scalars);
free(data.pubkeys);
+ free(data.pubkeys_gej);
free(data.seckeys);
free(data.output);
free(data.expected_output);