void run_benchmark(struct parameters p) {
init_benchmark(p);
benchmark.mapping_size = getpagesize() * MAPPING_SIZE;
benchmark.free_space = benchmark.mapping_size;
benchmark.max_deallocs = p.max_allocs;
benchmark.allocs_size = 0;
benchmark.allocs_number = 0;
benchmark.deallocs_number = 0;
benchmark.runs = 0;
benchmark.alloc_time = 0;
int previous_allocs = 0;
int previous_deallocs = 0;
bool alloc_result;
while(true) {
benchmark.runs++;
clock_t start_time = clock();
alloc_result = run_allocations(p);
clock_t end_time = clock();
benchmark.alloc_time += end_time - start_time;
if(!alloc_result && (benchmark.max_deallocs == 0 || !p.deallocate_after_fail))
break;
int old_max_deallocs = benchmark.max_deallocs;
run_deallocations(p);
if(DO_PRINT)
printf("Allocs: %d | Deallocs: %d | ",
benchmark.allocs_number - previous_allocs,
benchmark.deallocs_number - previous_deallocs);
if(DO_PRINT)
printf("Running diference: %d | Max deallocs: %d\n",
benchmark.allocs_number - benchmark.deallocs_number, old_max_deallocs);
previous_allocs = benchmark.allocs_number;
previous_deallocs = benchmark.deallocs_number;
}
save_result(p);
if(DO_PRINT) {
printf("TEST STOPPED\n");
printf("FULL DEALLOCATION\n");
}
run_full_deallocation();
}
void run_benchmark_1op(F f, OS& out, std::size_t size, std::size_t iter, init_method init = init_method::classic)
{
bench_vector<float> f_lhs, f_rhs, f_res;
bench_vector<double> d_lhs, d_rhs, d_res;
switch (init)
{
case init_method::classic:
init_benchmark(f_lhs, f_rhs, f_res, size);
init_benchmark(d_lhs, d_rhs, d_res, size);
break;
case init_method::arctrigo:
init_benchmark_arctrigo(f_lhs, f_rhs, f_res, size);
init_benchmark_arctrigo(d_lhs, d_rhs, d_res, size);
break;
default:
init_benchmark(f_lhs, f_rhs, f_res, size);
init_benchmark(d_lhs, d_rhs, d_res, size);
break;
}
duration_type t_float_scalar = benchmark_scalar(f, f_lhs, f_res, iter);
duration_type t_double_scalar = benchmark_scalar(f, d_lhs, d_res, iter);
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
duration_type t_float_sse = benchmark_simd<batch<float, 4>>(f, f_lhs, f_res, iter);
duration_type t_float_sse_u = benchmark_simd_unrolled<batch<float, 4>>(f, f_lhs, f_res, iter);
duration_type t_double_sse = benchmark_simd<batch<double, 2>>(f, d_lhs, d_res, iter);
duration_type t_double_sse_u = benchmark_simd_unrolled<batch<double, 2>>(f, d_lhs, d_res, iter);
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
duration_type t_float_avx = benchmark_simd<batch<float, 8>>(f, f_lhs, f_res, iter);
duration_type t_float_avx_u = benchmark_simd_unrolled<batch<float, 8>>(f, f_lhs, f_res, iter);
duration_type t_double_avx = benchmark_simd<batch<double, 4>>(f, d_lhs, d_res, iter);
duration_type t_double_avx_u = benchmark_simd_unrolled<batch<double, 4>>(f, d_lhs, d_res, iter);
#endif
#if defined(XSIMD_ARM_INSTR_SET)
duration_type t_float_neon = benchmark_simd<batch<float, 4>>(f, f_lhs, f_res, iter);
duration_type t_float_neon_u = benchmark_simd_unrolled<batch<float, 4>>(f, f_lhs, f_res, iter);
duration_type t_double_neon = benchmark_simd<batch<double, 2>>(f, d_lhs, d_res, iter);
duration_type t_double_neon_u = benchmark_simd_unrolled<batch<double, 2>>(f, d_lhs, d_res, iter);
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
duration_type t_float_fallback = benchmark_simd<batch<float, 7>>(f, f_lhs, f_res, iter);
duration_type t_float_fallback_u = benchmark_simd_unrolled<batch<float, 7>>(f, f_lhs, f_res, iter);
duration_type t_double_fallback = benchmark_simd<batch<double, 3>>(f, d_lhs, d_res, iter);
duration_type t_double_fallback_u = benchmark_simd_unrolled<batch<double, 3>>(f, d_lhs, d_res, iter);
#endif
out << "============================" << std::endl;
out << f.name() << std::endl;
out << "scalar float : " << t_float_scalar.count() << "ms" << std::endl;
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
out << "sse float : " << t_float_sse.count() << "ms" << std::endl;
out << "sse float unr : " << t_float_sse_u.count() << "ms" << std::endl;
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
out << "avx float : " << t_float_avx.count() << "ms" << std::endl;
out << "avx float unr : " << t_float_avx_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ARM_INSTR_SET)
out << "neon float : " << t_float_neon.count() << "ms" << std::endl;
out << "neon float unr : " << t_float_neon_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
out << "flbk float : " << t_float_fallback.count() << "ms" << std::endl;
out << "flbk float unr : " << t_float_fallback_u.count() << "ms" << std::endl;
#endif
out << "scalar double : " << t_double_scalar.count() << "ms" << std::endl;
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
out << "sse double : " << t_double_sse.count() << "ms" << std::endl;
out << "sse double unr : " << t_double_sse_u.count() << "ms" << std::endl;
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
out << "avx double : " << t_double_avx.count() << "ms" << std::endl;
out << "avx double unr : " << t_double_avx_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ARM_INSTR_SET)
out << "neon double : " << t_double_neon.count() << "ms" << std::endl;
out << "neon double unr: " << t_double_neon_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
out << "flbk double : " << t_double_fallback.count() << "ms" << std::endl;
out << "flbk double unr: " << t_double_fallback_u.count() << "ms" << std::endl;
#endif
out << "============================" << std::endl;
}
void run_benchmark_3op(F f, OS& out, std::size_t size, std::size_t iter)
{
bench_vector<float> f_op0, f_op1, f_op2, f_res;
bench_vector<double> d_op0, d_op1, d_op2, d_res;
init_benchmark(f_op0, f_op1, f_op2, f_res, size);
init_benchmark(d_op0, d_op1, d_op2, d_res, size);
duration_type t_float_scalar = benchmark_scalar(f, f_op0, f_op1, f_op2, f_res, iter);
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
duration_type t_float_sse = benchmark_simd<batch<float, 4>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_float_sse_u = benchmark_simd_unrolled<batch<float, 4>>(f, f_op0, f_op1, f_op2, f_res, iter);
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
duration_type t_float_avx = benchmark_simd<batch<float, 8>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_float_avx_u = benchmark_simd_unrolled<batch<float, 8>>(f, f_op0, f_op1, f_op2, f_res, iter);
#endif
duration_type t_double_scalar = benchmark_scalar(f, d_op0, d_op1, d_op2, d_res, iter);
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
duration_type t_double_sse = benchmark_simd<batch<double, 2>>(f, d_op0, d_op1, d_op2, d_res, iter);
duration_type t_double_sse_u = benchmark_simd_unrolled<batch<double, 2>>(f, d_op0, d_op1, d_op2, d_res, iter);
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
duration_type t_double_avx = benchmark_simd<batch<double, 4>>(f, d_op0, d_op1, d_op2, d_res, iter);
duration_type t_double_avx_u = benchmark_simd_unrolled<batch<double, 4>>(f, d_op0, d_op1, d_op2, d_res, iter);
#endif
#if defined(XSIMD_ARM_INSTR_SET)
duration_type t_float_neon = benchmark_simd<batch<float, 4>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_float_neon_u = benchmark_simd_unrolled<batch<float, 4>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_double_neon = benchmark_simd<batch<double, 2>>(f, d_op0, d_op1, d_op2, d_res, iter);
duration_type t_double_neon_u = benchmark_simd_unrolled<batch<double, 2>>(f, d_op0, d_op1, d_op2, d_res, iter);
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
duration_type t_float_fallback = benchmark_simd<batch<float, 7>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_float_fallback_u = benchmark_simd_unrolled<batch<float, 7>>(f, f_op0, f_op1, f_op2, f_res, iter);
duration_type t_double_fallback = benchmark_simd<batch<double, 3>>(f, d_op0, d_op1, d_op2, d_res, iter);
duration_type t_double_fallback_u = benchmark_simd_unrolled<batch<double, 3>>(f, d_op0, d_op1, d_op2, d_res, iter);
#endif
out << "============================" << std::endl;
out << f.name() << std::endl;
out << "scalar float : " << t_float_scalar.count() << "ms" << std::endl;
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
out << "sse float : " << t_float_sse.count() << "ms" << std::endl;
out << "sse float unr : " << t_float_sse_u.count() << "ms" << std::endl;
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
out << "avx float : " << t_float_avx.count() << "ms" << std::endl;
out << "avx float unr : " << t_float_avx_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ARM_INSTR_SET)
out << "neon float : " << t_float_neon.count() << "ms" << std::endl;
out << "neon float unr : " << t_float_neon_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
out << "flbk float : " << t_float_fallback.count() << "ms" << std::endl;
out << "flbk float unr : " << t_float_fallback_u.count() << "ms" << std::endl;
#endif
out << "scalar double : " << t_double_scalar.count() << "ms" << std::endl;
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_SSE2_VERSION
out << "sse double : " << t_double_sse.count() << "ms" << std::endl;
out << "sse double unr : " << t_double_sse_u.count() << "ms" << std::endl;
#endif
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX_VERSION
out << "avx double : " << t_double_avx.count() << "ms" << std::endl;
out << "avx double unr : " << t_double_avx_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ARM_INSTR_SET)
out << "neon double : " << t_double_neon.count() << "ms" << std::endl;
out << "neon double unr: " << t_double_neon_u.count() << "ms" << std::endl;
#endif
#if defined(XSIMD_ENABLE_FALLBACK)
out << "flbk double : " << t_double_fallback.count() << "ms" << std::endl;
out << "flbk double unr: " << t_double_fallback_u.count() << "ms" << std::endl;
#endif
out << "============================" << std::endl;
}
int main(int argc, char *argv[]) {
char *filename;
int32_t code;
FILE *dump;
double time, mem_used;
if (argc > 2) {
fprintf(stderr, "Usage: %s <filename>\n", argv[0]);
exit(YICES_EXIT_USAGE);
}
if (argc == 2) {
// read from file
filename = argv[1];
if (init_smt_file_lexer(&lexer, filename) < 0) {
perror(filename);
exit(YICES_EXIT_FILE_NOT_FOUND);
}
} else {
// read from stdin
init_smt_stdin_lexer(&lexer);
}
yices_init();
init_smt_tstack(&stack);
init_parser(&parser, &lexer, &stack);
init_benchmark(&bench);
code = parse_smt_benchmark(&parser, &bench);
if (code == 0) {
printf("No syntax error found\n");
}
if (benchmark_reduced_to_false(&bench)) {
printf("Reduced to false\n\nunsat\n");
fflush(stdout);
}
printf("\n");
time = get_cpu_time();
mem_used = mem_size() / (1024 * 1024);
printf("Construction time: %.4f s\n", time);
printf("Memory used: %.2f MB\n\n", mem_used);
fflush(stdout);
dump = fopen("yices2new.dmp", "w");
if (dump == NULL) {
perror("yices2new.dmp");
} else {
dump_benchmark(dump, &bench);
fclose(dump);
}
delete_benchmark(&bench);
delete_parser(&parser);
close_lexer(&lexer);
delete_tstack(&stack);
yices_exit();
return YICES_EXIT_SUCCESS;
}