X-Git-Url: https://sigrok.org/gitweb/?p=libsigrok.git;a=blobdiff_plain;f=tests%2Fanalog.c;h=8490bf9fca2a14e90cb4e04eb4f2f72fc5a9ddb3;hp=b5f2beb8e21bfd5df43f3090258b7d5c19fc72de;hb=4fad41a8a4f3;hpb=ee1b6054d6d6a00698ddae421eae63019c452b6d diff --git a/tests/analog.c b/tests/analog.c index b5f2beb8..8490bf9f 100644 --- a/tests/analog.c +++ b/tests/analog.c @@ -14,8 +14,7 @@ * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + * along with this program; if not, see . */ #include @@ -25,6 +24,36 @@ #include #include "lib.h" +/* + * This test sequence cannot use internal helpers, since it's limited + * to the library's public API (by design). That is why there are local + * helper routines for endianess handling. + */ + +static int host_be; + +static void get_host_endianess(void) +{ + int x; + uint8_t *p; + + p = (void *)&x; + x = 1; + host_be = *p ? 0 : 1; +} + +static void swap_bytes(uint8_t *buff, size_t blen) +{ + size_t idx; + uint8_t tmp; + + for (idx = 0; idx < blen / 2; idx++) { + tmp = buff[blen - 1 - idx]; + buff[blen - 1 - idx] = buff[idx]; + buff[idx] = tmp; + } +} + static int sr_analog_init_(struct sr_datafeed_analog *analog, struct sr_analog_encoding *encoding, struct sr_analog_meaning *meaning, @@ -124,6 +153,346 @@ START_TEST(test_analog_to_float_null) } END_TEST +START_TEST(test_analog_to_float_conv) +{ + static const int with_diag = 0; + + struct { + const char *desc; + void *bytes; + size_t nums, unit; + int is_fp, is_sign, is_be; + int scale, offset; + float *want; + } *item, items[] = { + /* Test to cover multiple values in an array, odd numbers. */ + { + .desc = "float single input, native, value array", + .bytes = (float[]){ -12.9, -333.999, 0, 3.14, 29.7, 9898.12, }, + .nums = 6, .unit = sizeof(float), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .scale = 1, .offset = 0, + .want = (float[]){ -12.9, -333.999, 0, 3.14, 29.7, 9898.12, }, + }, + /* Tests to cover floating point input data conversion. */ + { + .desc = "float single input, native", + .bytes = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + .nums = 4, .unit = sizeof(float), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "float single input, big endian", + .bytes = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + .nums = 4, .unit = sizeof(float), + .is_fp = TRUE, .is_sign = FALSE, .is_be = TRUE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "float single input, little endian", + .bytes = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + .nums = 4, .unit = sizeof(float), + .is_fp = TRUE, .is_sign = FALSE, .is_be = FALSE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "float double input, native", + .bytes = (double[]){ 1.0, 2.0, 3.0, 4.0, }, + .nums = 4, .unit = sizeof(double), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "float half input, unsupported, fake bytes", + .bytes = (uint16_t[]){ 0x1234, 0x5678, }, + .nums = 2, .unit = sizeof(uint16_t), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .want = NULL, + }, + { + .desc = "float quad input, unsupported, fake bytes", + .bytes = (uint64_t[]){ 0x0, 0x0, }, + .nums = 1, .unit = 2 * sizeof(uint64_t), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .want = NULL, + }, + /* Tests to cover integer input data conversion. */ + { + .desc = "int u8 input", + .bytes = (uint8_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint8_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = host_be, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "int i8 input", + .bytes = (int8_t[]){ -1, 2, -3, 4, }, + .nums = 4, .unit = sizeof(int8_t), + .is_fp = FALSE, .is_sign = TRUE, .is_be = host_be, + .scale = 1, .offset = 0, + .want = (float[]){ -1.0, 2.0, -3.0, 4.0, }, + }, + { + .desc = "int u16 input, big endian", + .bytes = (uint16_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint16_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = TRUE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "int u16 input, little endian", + .bytes = (uint16_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint16_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = FALSE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "int i16 input, big endian", + .bytes = (int16_t[]){ 1, -2, 3, -4, }, + .nums = 4, .unit = sizeof(int16_t), + .is_fp = FALSE, .is_sign = TRUE, .is_be = TRUE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, -2.0, 3.0, -4.0, }, + }, + { + .desc = "int i16 input, little endian", + .bytes = (int16_t[]){ 1, -2, 3, -4, }, + .nums = 4, .unit = sizeof(int16_t), + .is_fp = FALSE, .is_sign = TRUE, .is_be = FALSE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, -2.0, 3.0, -4.0, }, + }, + { + .desc = "int u32 input, big endian", + .bytes = (uint32_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint32_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = TRUE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "int u32 input, little endian", + .bytes = (uint32_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint32_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = FALSE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + }, + { + .desc = "int i32 input, big endian", + .bytes = (int32_t[]){ 1, 2, -3, -4, }, + .nums = 4, .unit = sizeof(int32_t), + .is_fp = FALSE, .is_sign = TRUE, .is_be = TRUE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, -3.0, -4.0, }, + }, + { + .desc = "int i32 input, little endian", + .bytes = (int32_t[]){ 1, 2, -3, -4, }, + .nums = 4, .unit = sizeof(int32_t), + .is_fp = FALSE, .is_sign = TRUE, .is_be = FALSE, + .scale = 1, .offset = 0, + .want = (float[]){ 1.0, 2.0, -3.0, -4.0, }, + }, + { + .desc = "int u64 input, unsupported", + .bytes = (uint64_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint64_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = TRUE, + .want = NULL, + }, + /* Tests to cover scale/offset calculation. */ + { + .desc = "float single input, scale + offset", + .bytes = (float[]){ 1.0, 2.0, 3.0, 4.0, }, + .nums = 4, .unit = sizeof(float), + .is_fp = TRUE, .is_sign = FALSE, .is_be = host_be, + .scale = 3, .offset = 2, + .want = (float[]){ 5.0, 8.0, 11.0, 14.0, }, + }, + { + .desc = "int u8 input, scale + offset", + .bytes = (uint8_t[]){ 1, 2, 3, 4, }, + .nums = 4, .unit = sizeof(uint8_t), + .is_fp = FALSE, .is_sign = FALSE, .is_be = TRUE, + .scale = 3, .offset = 2, + .want = (float[]){ 5.0, 8.0, 11.0, 14.0, }, + }, + }; + const size_t max_floats = 6; + struct sr_channel ch = { + .index = 0, + .enabled = TRUE, + .type = SR_CHANNEL_LOGIC, + .name = "input", + }; + + size_t item_idx; + char item_text[32]; + struct sr_datafeed_analog analog; + struct sr_analog_encoding encoding; + struct sr_analog_meaning meaning; + struct sr_analog_spec spec; + size_t byte_count, value_idx; + uint8_t f_in[max_floats * sizeof(double)], *byte_ptr; + float f_out[max_floats]; + int ret; + float want, have; + + for (item_idx = 0; item_idx < ARRAY_SIZE(items); item_idx++) { + item = &items[item_idx]; + + /* Construct "4x u32le" style test item identification. */ + snprintf(item_text, sizeof(item_text), "%zu: %zux %c%zu%s", + item_idx, item->nums, + item->is_fp ? 'f' : item->is_sign ? 'i' : 'u', + item->unit * 8, item->is_be ? "be" : "le"); + if (with_diag) { + fprintf(stderr, "%s -- %s", item_text, item->desc); + fflush(stderr); + } + + /* Copy input data bytes, optionally adjust endianess. */ + byte_count = item->nums * item->unit; + memcpy(f_in, item->bytes, byte_count); + if (item->is_be != host_be) { + byte_ptr = &f_in[0]; + for (value_idx = 0; value_idx < item->nums; value_idx++) { + swap_bytes(byte_ptr, item->unit); + byte_ptr += item->unit; + } + } + if (with_diag) { + fprintf(stderr, " -- bytes:"); + for (value_idx = 0; value_idx < byte_count; value_idx++) + fprintf(stderr, " %02x", f_in[value_idx]); + fflush(stderr); + } + + /* Setup the analog feed description. */ + sr_analog_init_(&analog, &encoding, &meaning, &spec, 3); + analog.num_samples = item->nums; + analog.data = &f_in[0]; + encoding.unitsize = item->unit; + encoding.is_float = item->is_fp; + encoding.is_signed = item->is_sign; + encoding.is_bigendian = item->is_be; + encoding.scale.p = item->scale ? item->scale : 1; + encoding.offset.p = item->offset; + meaning.channels = g_slist_append(NULL, &ch); + + /* Convert to an array of single precision float values. */ + ret = sr_analog_to_float(&analog, &f_out[0]); + if (!item->want) { + fail_if(ret == SR_OK, + "%s: sr_analog_to_float() passed", item_text); + if (with_diag) { + fprintf(stderr, " -- expected fail, OK\n"); + fflush(stderr); + } + continue; + } + fail_unless(ret == SR_OK, + "%s: sr_analog_to_float() failed: %d", item_text, ret); + if (with_diag) { + fprintf(stderr, " -- float:"); + for (value_idx = 0; value_idx < item->nums; value_idx++) + fprintf(stderr, " %f", f_out[value_idx]); + fprintf(stderr, "\n"); + fflush(stderr); + } + + /* + * Compare result data to the expectation. No tolerance + * is required here due to the input set's values. This + * test concentrates on endianess / data type / bit count + * conversion and simple scale/offset calculation, neither + * on precision nor rounding nor truncation. + */ + for (value_idx = 0; value_idx < item->nums; value_idx++) { + want = item->want[value_idx]; + have = f_out[value_idx]; + fail_unless(want == have, + "%s: input %f != output %f", + item_text, want, have); + } + } +} +END_TEST + +START_TEST(test_analog_si_prefix) +{ + struct { + float input_value; + int input_digits; + float output_value; + int output_digits; + const char *output_si_prefix; + } v[] = { + { 12.0 , 0, 12.0 , 0, "" }, + { 12.0 , 1, 12.0 , 1, "" }, + { 12.0 , -1, 0.012, 2, "k" }, + { 1024.0 , 0, 1.024, 3, "k" }, + { 1024.0 , -1, 1.024, 2, "k" }, + { 1024.0 , -3, 1.024, 0, "k" }, + { 12.0e5 , 0, 1.2, 6, "M" }, + { 0.123456, 0, 0.123456, 0, "" }, + { 0.123456, 1, 0.123456, 1, "" }, + { 0.123456, 2, 0.123456, 2, "" }, + { 0.123456, 3, 123.456, 0, "m" }, + { 0.123456, 4, 123.456, 1, "m" }, + { 0.123456, 5, 123.456, 2, "m" }, + { 0.123456, 6, 123.456, 3, "m" }, + { 0.123456, 7, 123.456, 4, "m" }, + { 0.0123 , 4, 12.3, 1, "m" }, + { 0.00123 , 5, 1.23, 2, "m" }, + { 0.000123, 4, 0.123, 1, "m" }, + { 0.000123, 5, 0.123, 2, "m" }, + { 0.000123, 6, 123.0, 0, "µ" }, + { 0.000123, 7, 123.0, 1, "µ" }, + }; + + for (unsigned int i = 0; i < ARRAY_SIZE(v); i++) { + float value = v[i].input_value; + int digits = v[i].input_digits; + const char *si_prefix = sr_analog_si_prefix(&value, &digits); + + fail_unless(fabs(value - v[i].output_value) <= 0.00001, + "sr_analog_si_prefix() unexpected output value %f (i=%d).", + value , i); + fail_unless(digits == v[i].output_digits, + "sr_analog_si_prefix() unexpected output digits %d (i=%d).", + digits, i); + fail_unless(!strcmp(si_prefix, v[i].output_si_prefix), + "sr_analog_si_prefix() unexpected output prefix \"%s\" (i=%d).", + si_prefix, i); + } +} +END_TEST + +START_TEST(test_analog_si_prefix_null) +{ + float value = 1.23; + int digits = 1; + const char *si_prefix; + + si_prefix = sr_analog_si_prefix(NULL, &digits); + fail_unless(!strcmp(si_prefix, "")); + si_prefix = sr_analog_si_prefix(&value, NULL); + fail_unless(!strcmp(si_prefix, "")); + si_prefix = sr_analog_si_prefix(NULL, NULL); + fail_unless(!strcmp(si_prefix, "")); +} +END_TEST + START_TEST(test_analog_unit_to_string) { int ret; @@ -133,13 +502,15 @@ START_TEST(test_analog_unit_to_string) struct sr_analog_encoding encoding; struct sr_analog_meaning meaning; struct sr_analog_spec spec; - const char *r[] = {" V RMS"}; + const int u[] = {SR_UNIT_VOLT, SR_UNIT_AMPERE, SR_UNIT_CELSIUS}; + const int f[] = {SR_MQFLAG_RMS, 0, 0}; + const char *r[] = {"V RMS", "A", "°C"}; sr_analog_init_(&analog, &encoding, &meaning, &spec, 3); - for (i = -1; i < ARRAY_SIZE(r); i++) { - meaning.unit = SR_UNIT_VOLT; - meaning.mqflags = SR_MQFLAG_RMS; + for (i = 0; i < ARRAY_SIZE(r); i++) { + meaning.unit = u[i]; + meaning.mqflags = f[i]; ret = sr_analog_unit_to_string(&analog, &result); fail_unless(ret == SR_OK); fail_unless(result != NULL); @@ -266,22 +637,78 @@ START_TEST(test_mult_rational) } END_TEST +START_TEST(test_div_rational) +{ + const struct sr_rational r[][3] = { + /* a * b = c */ + { { 1, 1 }, { 1, 1 }, { 1, 1 }}, + { { 2, 1 }, { 1, 3 }, { 6, 1 }}, + { { 1, 2 }, { 1, 2 }, { 1, 1 }}, + /* Test negative numbers */ + { { -1, 2 }, { 1, 2 }, { -1, 1 }}, + { { -1, 2 }, { -1, 2 }, { 1, 1 }}, + { { -(1ll<<20), (1ll<<10) }, { -1, (1ll<<20) }, { (1ll<<30), 1 }}, + /* Test reduction */ + { { INT32_MAX, (1ll<<12) }, { 1, (1<<2) }, { INT32_MAX, (1ll<<10) }}, + { { INT64_MAX, (1ll<<63) }, { 1, (1<<3) }, { INT64_MAX, (1ll<<60) }}, + /* Test large numbers */ + { { (1ll<<40), (1ll<<10) }, { 1, (1ll<<30) }, { (1ll<<60), 1 }}, + { { -(1ll<<40), (1ll<<10) }, { -1, (1ll<<30) }, { (1ll<<60), 1 }}, + + { { 10000*3, 4 }, { 1, 80000*3 }, { 200000000*9, 1 }}, + { { 4, 10000*3 }, { 80000*3, 1 }, { 1, 200000000*9 }}, + + { { -10000*3, 4 }, { 1, 80000*3 }, { -200000000*9, 1 }}, + { { 10000*3, 4 }, { -1, 80000*3 }, { -200000000*9, 1 }}, + }; + + for (unsigned i = 0; i < ARRAY_SIZE(r); i++) { + struct sr_rational res; + + int rc = sr_rational_div(&res, &r[i][0], &r[i][1]); + fail_unless(rc == SR_OK); + fail_unless(sr_rational_eq(&res, &r[i][2]) == 1, + "sr_rational_mult() failed: [%d] %ld/%lu != %ld/%lu.", + i, res.p, res.q, r[i][2].p, r[i][2].q); + } + + { + struct sr_rational res; + int rc = sr_rational_div(&res, &r[0][0], &((struct sr_rational){ 0, 5 })); + + fail_unless(rc == SR_ERR_ARG); + } +} +END_TEST + Suite *suite_analog(void) { Suite *s; TCase *tc; + get_host_endianess(); + s = suite_create("analog"); tc = tcase_create("analog_to_float"); tcase_add_test(tc, test_analog_to_float); tcase_add_test(tc, test_analog_to_float_null); + tcase_add_test(tc, test_analog_to_float_conv); + suite_add_tcase(s, tc); + + tc = tcase_create("analog_si_unit"); + tcase_add_test(tc, test_analog_si_prefix); + tcase_add_test(tc, test_analog_si_prefix_null); tcase_add_test(tc, test_analog_unit_to_string); tcase_add_test(tc, test_analog_unit_to_string_null); + suite_add_tcase(s, tc); + + tc = tcase_create("analog_rational"); tcase_add_test(tc, test_set_rational); tcase_add_test(tc, test_set_rational_null); tcase_add_test(tc, test_cmp_rational); tcase_add_test(tc, test_mult_rational); + tcase_add_test(tc, test_div_rational); suite_add_tcase(s, tc); return s;