[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAGS_qxrfTeW=TmrbAUoUw5AOMzq+YccV3hTL6TY9b5+3OryWyw@mail.gmail.com>
Date: Thu, 2 Sep 2021 19:26:17 -0700
From: Daniel Latypov <dlatypov@...gle.com>
To: andriy.shevchenko@...ux.intel.com
Cc: brendanhiggins@...gle.com, davidgow@...gle.com,
linux-kernel@...r.kernel.org, kunit-dev@...glegroups.com,
linux-kselftest@...r.kernel.org, skhan@...uxfoundation.org
Subject: Re: [PATCH v5] lib: add basic KUnit test for lib/math
On Mon, Apr 12, 2021 at 12:07 PM Daniel Latypov <dlatypov@...gle.com> wrote:
>
> Add basic test coverage for files that don't require any config options:
> * part of math.h (what seem to be the most commonly used macros)
> * gcd.c
> * lcm.c
> * int_sqrt.c
> * reciprocal_div.c
> (Ignored int_pow.c since it's a simple textbook algorithm.)
>
> These tests aren't particularly interesting, but they
> * provide short and simple examples of parameterized tests
> * provide a place to add tests for any new files in this dir
> * are written so adding new test cases to cover edge cases should be easy
> * looking at code coverage, we hit all the branches in the .c files
Bumping this thread to see what the status is.
I think it's useful to have tests spread across the kernel so there's
"nearby" tests one can reference and/or copy-paste from.
Now there's the lib/math/rational-test.c, there's less need here.
But I think having tests for simpler functions is still nice to have.
E.g. there's a performance trade-off documented in gcd.c
If you'd want to run that test case to see if it still holds, you'd be
able to run gcd() in a loop fairly easily with:
$ ./tools/testing/kunit/kunit.py run --kunitconfig=lib/math lib-math.gcd_test
(KUnit doesn't yet have support for timing tests or running tests
multiple times, so you'd have to tweak the test code for that).
Random fact:
I'd noticed that after running some 1000s of internal integration
tests , the gcd() function didn't actually get fully covered.
Specifically, they never got to line 34, the for loop
23 unsigned long gcd(unsigned long a, unsigned long b)
24 {
25 unsigned long r = a | b;
26
27 if (!a || !b)
28 return r;
29
30 b >>= __ffs(b);
31 if (b == 1)
32 return r & -r;
33
34 for (;;) {
Checking again now, a few months later, I see they now do hit that loop.
But I guess until then, we'd only been calling gcd() with b=2^n.
>
> Signed-off-by: Daniel Latypov <dlatypov@...gle.com>
> ---
> Changes since v4:
> * add in test cases for some math.h macros (abs, round_up/round_down,
> div_round_down/closest)
> * use parameterized testing less to keep things terser
>
> Changes since v3:
> * fix `checkpatch.pl --strict` warnings
> * add test cases for gcd(0,0) and lcm(0,0)
> * minor: don't test both gcd(a,b) and gcd(b,a) when a == b
>
> Changes since v2: mv math_test.c => math_kunit.c
>
> Changes since v1:
> * Rebase and rewrite to use the new parameterized testing support.
> * misc: fix overflow in literal and inline int_sqrt format string.
> * related: commit 1f0e943df68a ("Documentation: kunit: provide guidance
> for testing many inputs") was merged explaining the patterns shown here.
> * there's an in-flight patch to update it for parameterized testing.
>
> v1: https://lore.kernel.org/lkml/20201019224556.3536790-1-dlatypov@google.com/
> ---
> lib/math/Kconfig | 5 +
> lib/math/Makefile | 2 +
> lib/math/math_kunit.c | 264 ++++++++++++++++++++++++++++++++++++++++++
> 3 files changed, 271 insertions(+)
> create mode 100644 lib/math/math_kunit.c
>
> diff --git a/lib/math/Kconfig b/lib/math/Kconfig
> index f19bc9734fa7..6ba8680439c1 100644
> --- a/lib/math/Kconfig
> +++ b/lib/math/Kconfig
> @@ -15,3 +15,8 @@ config PRIME_NUMBERS
>
> config RATIONAL
> bool
> +
> +config MATH_KUNIT_TEST
> + tristate "KUnit test for lib/math" if !KUNIT_ALL_TESTS
> + default KUNIT_ALL_TESTS
> + depends on KUNIT
> diff --git a/lib/math/Makefile b/lib/math/Makefile
> index be6909e943bd..30abb7a8d564 100644
> --- a/lib/math/Makefile
> +++ b/lib/math/Makefile
> @@ -4,3 +4,5 @@ obj-y += div64.o gcd.o lcm.o int_pow.o int_sqrt.o reciprocal_div.o
> obj-$(CONFIG_CORDIC) += cordic.o
> obj-$(CONFIG_PRIME_NUMBERS) += prime_numbers.o
> obj-$(CONFIG_RATIONAL) += rational.o
> +
> +obj-$(CONFIG_MATH_KUNIT_TEST) += math_kunit.o
> diff --git a/lib/math/math_kunit.c b/lib/math/math_kunit.c
> new file mode 100644
> index 000000000000..80a087a32884
> --- /dev/null
> +++ b/lib/math/math_kunit.c
> @@ -0,0 +1,264 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Simple KUnit suite for math helper funcs that are always enabled.
> + *
> + * Copyright (C) 2020, Google LLC.
> + * Author: Daniel Latypov <dlatypov@...gle.com>
> + */
> +
> +#include <kunit/test.h>
> +#include <linux/gcd.h>
> +#include <linux/kernel.h>
> +#include <linux/lcm.h>
> +#include <linux/reciprocal_div.h>
> +
> +static void abs_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, abs('\0'), '\0');
> + KUNIT_EXPECT_EQ(test, abs('a'), 'a');
> + KUNIT_EXPECT_EQ(test, abs(-'a'), 'a');
> +
> + /* The expression in the macro is actually promoted to an int. */
> + KUNIT_EXPECT_EQ(test, abs((short)0), 0);
> + KUNIT_EXPECT_EQ(test, abs((short)42), 42);
> + KUNIT_EXPECT_EQ(test, abs((short)-42), 42);
> +
> + KUNIT_EXPECT_EQ(test, abs(0), 0);
> + KUNIT_EXPECT_EQ(test, abs(42), 42);
> + KUNIT_EXPECT_EQ(test, abs(-42), 42);
> +
> + KUNIT_EXPECT_EQ(test, abs(0L), 0L);
> + KUNIT_EXPECT_EQ(test, abs(42L), 42L);
> + KUNIT_EXPECT_EQ(test, abs(-42L), 42L);
> +
> + KUNIT_EXPECT_EQ(test, abs(0LL), 0LL);
> + KUNIT_EXPECT_EQ(test, abs(42LL), 42LL);
> + KUNIT_EXPECT_EQ(test, abs(-42LL), 42LL);
> +
> + /* Unsigned types get casted to signed. */
> + KUNIT_EXPECT_EQ(test, abs(0ULL), 0LL);
> + KUNIT_EXPECT_EQ(test, abs(42ULL), 42LL);
> +}
> +
> +static void int_sqrt_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, int_sqrt(0UL), 0UL);
> + KUNIT_EXPECT_EQ(test, int_sqrt(1UL), 1UL);
> + KUNIT_EXPECT_EQ(test, int_sqrt(4UL), 2UL);
> + KUNIT_EXPECT_EQ(test, int_sqrt(5UL), 2UL);
> + KUNIT_EXPECT_EQ(test, int_sqrt(8UL), 2UL);
> + KUNIT_EXPECT_EQ(test, int_sqrt(1UL << 30), 1UL << 15);
> +}
> +
> +static void round_up_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, round_up(0, 1), 0);
> + KUNIT_EXPECT_EQ(test, round_up(1, 2), 2);
> + KUNIT_EXPECT_EQ(test, round_up(3, 2), 4);
> + KUNIT_EXPECT_EQ(test, round_up((1 << 30) - 1, 2), 1 << 30);
> + KUNIT_EXPECT_EQ(test, round_up((1 << 30) - 1, 1 << 29), 1 << 30);
> +}
> +
> +static void round_down_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, round_down(0, 1), 0);
> + KUNIT_EXPECT_EQ(test, round_down(1, 2), 0);
> + KUNIT_EXPECT_EQ(test, round_down(3, 2), 2);
> + KUNIT_EXPECT_EQ(test, round_down((1 << 30) - 1, 2), (1 << 30) - 2);
> + KUNIT_EXPECT_EQ(test, round_down((1 << 30) - 1, 1 << 29), 1 << 29);
> +}
> +
> +static void div_round_up_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_UP(0, 1), 0);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_UP(20, 10), 2);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_UP(21, 10), 3);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_UP(21, 20), 2);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_UP(21, 99), 1);
> +}
> +
> +static void div_round_closest_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_CLOSEST(0, 1), 0);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_CLOSEST(20, 10), 2);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_CLOSEST(21, 10), 2);
> + KUNIT_EXPECT_EQ(test, DIV_ROUND_CLOSEST(25, 10), 3);
> +}
> +
> +/* Generic test case for unsigned long inputs. */
> +struct test_case {
> + unsigned long a, b;
> + unsigned long result;
> +};
> +
> +static struct test_case gcd_cases[] = {
> + {
> + .a = 0, .b = 0,
> + .result = 0,
> + },
> + {
> + .a = 0, .b = 1,
> + .result = 1,
> + },
> + {
> + .a = 2, .b = 2,
> + .result = 2,
> + },
> + {
> + .a = 2, .b = 4,
> + .result = 2,
> + },
> + {
> + .a = 3, .b = 5,
> + .result = 1,
> + },
> + {
> + .a = 3 * 9, .b = 3 * 5,
> + .result = 3,
> + },
> + {
> + .a = 3 * 5 * 7, .b = 3 * 5 * 11,
> + .result = 15,
> + },
> + {
> + .a = 1 << 21,
> + .b = (1 << 21) - 1,
> + .result = 1,
> + },
> +};
> +
> +KUNIT_ARRAY_PARAM(gcd, gcd_cases, NULL);
> +
> +static void gcd_test(struct kunit *test)
> +{
> + const char *message_fmt = "gcd(%lu, %lu)";
> + const struct test_case *test_param = test->param_value;
> +
> + KUNIT_EXPECT_EQ_MSG(test, test_param->result,
> + gcd(test_param->a, test_param->b),
> + message_fmt, test_param->a,
> + test_param->b);
> +
> + if (test_param->a == test_param->b)
> + return;
> +
> + /* gcd(a,b) == gcd(b,a) */
> + KUNIT_EXPECT_EQ_MSG(test, test_param->result,
> + gcd(test_param->b, test_param->a),
> + message_fmt, test_param->b,
> + test_param->a);
> +}
> +
> +static struct test_case lcm_cases[] = {
> + {
> + .a = 0, .b = 0,
> + .result = 0,
> + },
> + {
> + .a = 0, .b = 1,
> + .result = 0,
> + },
> + {
> + .a = 1, .b = 2,
> + .result = 2,
> + },
> + {
> + .a = 2, .b = 2,
> + .result = 2,
> + },
> + {
> + .a = 3 * 5, .b = 3 * 7,
> + .result = 3 * 5 * 7,
> + },
> +};
> +
> +KUNIT_ARRAY_PARAM(lcm, lcm_cases, NULL);
> +
> +static void lcm_test(struct kunit *test)
> +{
> + const char *message_fmt = "lcm(%lu, %lu)";
> + const struct test_case *test_param = test->param_value;
> +
> + KUNIT_EXPECT_EQ_MSG(test, test_param->result,
> + lcm(test_param->a, test_param->b),
> + message_fmt, test_param->a,
> + test_param->b);
> +
> + if (test_param->a == test_param->b)
> + return;
> +
> + /* lcm(a,b) == lcm(b,a) */
> + KUNIT_EXPECT_EQ_MSG(test, test_param->result,
> + lcm(test_param->b, test_param->a),
> + message_fmt, test_param->b,
> + test_param->a);
> +}
> +
> +struct u32_test_case {
> + u32 a, b;
> + u32 result;
> +};
> +
> +static struct u32_test_case reciprocal_div_cases[] = {
> + {
> + .a = 0, .b = 1,
> + .result = 0,
> + },
> + {
> + .a = 42, .b = 20,
> + .result = 2,
> + },
> + {
> + .a = 42, .b = 9999,
> + .result = 0,
> + },
> + {
> + .a = (1 << 16), .b = (1 << 14),
> + .result = 1 << 2,
> + },
> +};
> +
> +KUNIT_ARRAY_PARAM(reciprocal_div, reciprocal_div_cases, NULL);
> +
> +static void reciprocal_div_test(struct kunit *test)
> +{
> + const struct u32_test_case *test_param = test->param_value;
> + struct reciprocal_value rv = reciprocal_value(test_param->b);
> +
> + KUNIT_EXPECT_EQ_MSG(test, test_param->result,
> + reciprocal_divide(test_param->a, rv),
> + "reciprocal_divide(%u, %u)",
> + test_param->a, test_param->b);
> +}
> +
> +static void reciprocal_scale_test(struct kunit *test)
> +{
> + KUNIT_EXPECT_EQ(test, reciprocal_scale(0u, 100), 0u);
> + KUNIT_EXPECT_EQ(test, reciprocal_scale(1u, 100), 0u);
> + KUNIT_EXPECT_EQ(test, reciprocal_scale(1u << 4, 1 << 28), 1u);
> + KUNIT_EXPECT_EQ(test, reciprocal_scale(1u << 16, 1 << 28), 1u << 12);
> + KUNIT_EXPECT_EQ(test, reciprocal_scale(~0u, 1 << 28), (1u << 28) - 1);
> +}
> +
> +static struct kunit_case math_test_cases[] = {
> + KUNIT_CASE(abs_test),
> + KUNIT_CASE(int_sqrt_test),
> + KUNIT_CASE(round_up_test),
> + KUNIT_CASE(round_down_test),
> + KUNIT_CASE(div_round_up_test),
> + KUNIT_CASE(div_round_closest_test),
> + KUNIT_CASE_PARAM(gcd_test, gcd_gen_params),
> + KUNIT_CASE_PARAM(lcm_test, lcm_gen_params),
> + KUNIT_CASE_PARAM(reciprocal_div_test, reciprocal_div_gen_params),
> + KUNIT_CASE(reciprocal_scale_test),
> + {}
> +};
> +
> +static struct kunit_suite math_test_suite = {
> + .name = "lib-math",
> + .test_cases = math_test_cases,
> +};
> +
> +kunit_test_suites(&math_test_suite);
> +
> +MODULE_LICENSE("GPL v2");
>
> base-commit: 4fa56ad0d12e24df768c98bffe9039f915d1bc02
> --
> 2.31.1.295.g9ea45b61b8-goog
>
Powered by blists - more mailing lists