Build the benchmarks as part of the CI (#5965)

* Build the benchmarks as part of 'linux-meson-gcc-tests' step
* Fix bench_il build error
* Add macro to avoid compiler optimization on benchmarked code
* Fix incorrect format specifier
* Cast to uint64_t

.github/workflows/ci.yml

@@ -82,7 +82,7 @@ jobs:
             build_system: meson
             compiler: gcc
             run_tests: true
-            meson_options: -Dbuildtype=release --werror
+            meson_options: -Dbuildtype=release -Denable_benchmarks=true --werror
             enabled: ${{ needs.changes.outputs.edited == 'true' }}
             timeout: 45
             cflags: -DRZ_ASSERT_STDOUT=1

test/bench/bench_ht.c

@@ -152,23 +152,23 @@ static void bench_rz_ht_pu_combined(RzTable *t_out) {
 		});
 		RZ_BENCH_RUN_I("[HtPU] lookup (100 elements)", i, t_out, ITERATION_COUNT, {
 			PUKey temp_key = make_pu_key(reshuffle_key(i, 100, UT64_MAX));
-			ut64 result = ht_pu_find(ht_100, &temp_key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_pu_find(ht_100, &temp_key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtPU] lookup (1k elements)", i, t_out, ITERATION_COUNT, {
 			PUKey temp_key = make_pu_key(reshuffle_key(i, 1000, UT64_MAX));
-			ut64 result = ht_pu_find(ht_1k, &temp_key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_pu_find(ht_1k, &temp_key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtPU] lookup (10k elements)", i, t_out, ITERATION_COUNT, {
 			PUKey temp_key = make_pu_key(reshuffle_key(i, 10000, UT64_MAX));
-			ut64 result = ht_pu_find(ht_10k, &temp_key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_pu_find(ht_10k, &temp_key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtPU] lookup (100k elements)", i, t_out, ITERATION_COUNT, {
 			PUKey temp_key = make_pu_key(reshuffle_key(i, 100000, UT64_MAX));
-			ut64 result = ht_pu_find(ht_100k, &temp_key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_pu_find(ht_100k, &temp_key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtPU] lookup (1M elements)", i, t_out, ITERATION_COUNT, {
 			PUKey temp_key = make_pu_key(reshuffle_key(i, 1000000, UT64_MAX));
-			ut64 result = ht_pu_find(ht_1m, &temp_key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_pu_find(ht_1m, &temp_key, NULL));
 		});
 
 		ht_pu_free(ht_100);
@@ -181,9 +181,12 @@ static void bench_rz_ht_pu_combined(RzTable *t_out) {
 	free(keys);
 }
 
-static char *generate_su_key(ut64 index) {
+static char *generate_su_key(ut64 i) {
 	char buffer[UT8_MAX];
 
+	// cast to uint64_t to avoid format specifier (PRIx64) mismatch
+	uint64_t index = (uint64_t)i;
+
 	// Try to mimic real world string keys
 	switch (index % 8) {
 	case 0:
@@ -296,23 +299,23 @@ static void bench_rz_ht_su_combined(RzTable *t_out) {
 		});
 		RZ_BENCH_RUN_I("[HtSU] lookup (100 elements)", i, t_out, ITERATION_COUNT, {
 			const char *key = precomputed_keys[reshuffle_key(i, 100, ITERATION_COUNT)];
-			ut64 result = ht_su_find(ht_100, key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_su_find(ht_100, key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtSU] lookup (1k elements)", i, t_out, ITERATION_COUNT, {
 			const char *key = precomputed_keys[reshuffle_key(i, 1000, ITERATION_COUNT)];
-			ut64 result = ht_su_find(ht_1k, key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_su_find(ht_1k, key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtSU] lookup (10k elements)", i, t_out, ITERATION_COUNT, {
 			const char *key = precomputed_keys[reshuffle_key(i, 10000, ITERATION_COUNT)];
-			ut64 result = ht_su_find(ht_10k, key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_su_find(ht_10k, key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtSU] lookup (100k elements)", i, t_out, ITERATION_COUNT, {
 			const char *key = precomputed_keys[reshuffle_key(i, 100000, ITERATION_COUNT)];
-			ut64 result = ht_su_find(ht_100k, key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_su_find(ht_100k, key, NULL));
 		});
 		RZ_BENCH_RUN_I("[HtSU] lookup (1M elements)", i, t_out, ITERATION_COUNT, {
 			const char *key = precomputed_keys[reshuffle_key(i, 1000000, ITERATION_COUNT)];
-			ut64 result = ht_su_find(ht_1m, key, NULL);
+			RZ_DONT_OPTIMIZE(ut64, ht_su_find(ht_1m, key, NULL));
 		});
 
 		ht_su_free(ht_100);

test/bench/bench_il.c

@@ -12,7 +12,7 @@
 
 static void bench_rz_il_mem_loadw(const char *title, RzTable *t_out, ut32 bit_size, bool big_endian) {
 	RzBuffer *buf = rz_buf_new_empty(bit_size);
-	RzILMem *mem = rz_il_mem_new(buf, 64);
+	RzILMem *mem = rz_il_mem_new_borrowed(buf, 64);
 	RzBitVector *key = rz_bv_new_from_ut64(64, 0x12345678abcdef);
 
 	RZ_BENCH_RUN(title, t_out, 1000000, {

test/bench/bench_util.c

@@ -13,7 +13,7 @@ static void bench_rz_bits_trailing_zeros(RzTable *t_out) {
 	ut64 v = 0;
 
 	RZ_BENCH_RUN("bench_rz_bits_trailing_zeros", t_out, 5000000, {
-		volatile ut64 result = rz_bits_trailing_zeros(v++);
+		RZ_DONT_OPTIMIZE(size_t, rz_bits_trailing_zeros(v++));
 	});
 }
 

test/bench/bench_utils.h

@@ -73,4 +73,33 @@ RZ_API void rz_bench_report(RZ_NONNULL RzBenchCtx *ctx, RZ_NONNULL RzTable *t);
 	free(table_out); \
 	rz_table_free(T);
 
+#if defined(__GNUC__) || defined(__clang__)
+/**
+ * \def Helper macro to prevent compiler optimizations on benchmarked code.
+ * \param type Return type of the expression \p x.
+ * \param x Function or code to be executed.
+ */
+#define RZ_DONT_OPTIMIZE(type, x) \
+	do { \
+		type tmp_ = (x); \
+		__asm__ volatile("" : : "r,m"(tmp_) : "memory"); \
+	} while (0)
+#elif defined(_MSC_VER)
+#include <intrin.h>
+#define RZ_DONT_OPTIMIZE(type, x) \
+	do { \
+		type tmp_ = (x); \
+		_WriteBarrier(); \
+		*(volatile char *)&tmp_; \
+		_ReadBarrier(); \
+	} while (0)
+#else
+// Fallback
+#define RZ_DONT_OPTIMIZE(type, x) \
+	do { \
+		volatile type tmp_ = (x); \
+		(void)tmp_; \
+	} while (0)
+#endif
+
 #endif // BENCH_UTILS_H

test/bench/meson.build

@@ -21,7 +21,7 @@ if get_option('enable_benchmarks')
   benchmarks = [
     'bitvector',
     'il',
-    'ht'
+    'ht',
     'util'
   ]