diff options
author | Jason Ekstrand <[email protected]> | 2018-09-13 09:56:20 -0500 |
---|---|---|
committer | Jason Ekstrand <[email protected]> | 2018-10-10 13:13:12 -0500 |
commit | 328d4d080bce128ff422fbfbd7271ef4ede38320 (patch) | |
tree | 6e557472267ec1a2cfa0b69122c76b9a7c5931cc /src/util/tests | |
parent | a9be8dddfedb1d19e43b900bdfd33731d3c390c4 (diff) |
util: Add tests for fast integer division by constants
While I generally trust rediculousfish to have done his homework, we've
made some adjustments to suit the needs of mesa and it'd be good to
test those. Also, there's no better place than unit tests to clearly
document the different edge cases of the different methods.
Reviewed-by: Marek Olšák <[email protected]>
Diffstat (limited to 'src/util/tests')
-rw-r--r-- | src/util/tests/fast_idiv_by_const/Makefile.am | 43 | ||||
-rw-r--r-- | src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp | 472 | ||||
-rw-r--r-- | src/util/tests/fast_idiv_by_const/meson.build | 30 |
3 files changed, 545 insertions, 0 deletions
diff --git a/src/util/tests/fast_idiv_by_const/Makefile.am b/src/util/tests/fast_idiv_by_const/Makefile.am new file mode 100644 index 00000000000..1ebee09f59b --- /dev/null +++ b/src/util/tests/fast_idiv_by_const/Makefile.am @@ -0,0 +1,43 @@ +# Copyright © 2018 Intel +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. + +AM_CPPFLAGS = \ + -I$(top_srcdir)/src \ + -I$(top_srcdir)/include \ + -I$(top_srcdir)/src/gallium/include \ + -I$(top_srcdir)/src/gtest/include \ + $(PTHREAD_CFLAGS) \ + $(DEFINES) + +TESTS = fast_idiv_by_const_test + +check_PROGRAMS = $(TESTS) + +fast_idiv_by_const_test_SOURCES = \ + fast_idiv_by_const_test.cpp + +fast_idiv_by_const_test_LDADD = \ + $(top_builddir)/src/gtest/libgtest.la \ + $(top_builddir)/src/util/libmesautil.la \ + $(PTHREAD_LIBS) \ + $(DLOPEN_LIBS) + +EXTRA_DIST = meson.build diff --git a/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp b/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp new file mode 100644 index 00000000000..3983a39edda --- /dev/null +++ b/src/util/tests/fast_idiv_by_const/fast_idiv_by_const_test.cpp @@ -0,0 +1,472 @@ +/* + * Copyright © 2018 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + */ + +#include <gtest/gtest.h> +#include "util/bigmath.h" +#include "util/fast_idiv_by_const.h" +#include "util/u_math.h" + +#define RAND_TEST_ITERATIONS 100000 + +#define MAX_UINT(bits) \ + (((bits) == 64) ? UINT64_MAX : ((1ull << (bits)) - 1)) + +static inline uint64_t +utrunc(uint64_t x, unsigned num_bits) +{ + if (num_bits == 64) + return x; + + return (x << (64 - num_bits)) >> (64 - num_bits); +} + +static inline int64_t +strunc(int64_t x, unsigned num_bits) +{ + if (num_bits == 64) + return x; + + return (x << (64 - num_bits)) >> (64 - num_bits); +} + +static inline bool +uint_is_in_range(uint64_t x, unsigned num_bits) +{ + if (num_bits == 64) + return true; + + return x < (1ull << num_bits); +} + +static inline bool +sint_is_in_range(int64_t x, unsigned num_bits) +{ + if (num_bits == 64) + return true; + + return x >= -(1ll << (num_bits - 1)) && + x < (1ll << (num_bits - 1)); +} + +static inline uint64_t +uadd_sat(uint64_t a, uint64_t b, unsigned num_bits) +{ + assert(uint_is_in_range(a, num_bits)); + assert(uint_is_in_range(b, num_bits)); + + uint64_t sum = a + b; + if (num_bits == 64) { + /* Standard overflow check */ + return sum < a ? UINT64_MAX : sum; + } else { + /* Check if sum is more than num_bits */ + return (sum >> num_bits) ? MAX_UINT(num_bits) : sum; + } +} + +static inline uint64_t +umul_add_high(uint64_t a, uint64_t b, uint64_t c, unsigned num_bits) +{ + assert(uint_is_in_range(a, num_bits)); + assert(uint_is_in_range(b, num_bits)); + assert(uint_is_in_range(c, num_bits)); + + if (num_bits == 64) { + uint32_t a32[2] = { (uint32_t)a, (uint32_t)(a >> 32) }; + uint32_t b32[2] = { (uint32_t)b, (uint32_t)(b >> 32) }; + uint32_t c32[2] = { (uint32_t)c, (uint32_t)(c >> 32) }; + + uint32_t ab32[4]; + ubm_mul_u32arr(ab32, a32, b32); + + uint32_t abc32[4]; + ubm_add_u32arr(abc32, ab32, c32); + + return abc32[2] | ((uint64_t)abc32[3] << 32); + } else { + assert(num_bits <= 32); + return utrunc(((a * b) + c) >> num_bits, num_bits); + } +} + +static inline int64_t +smul_high(int64_t a, int64_t b, unsigned num_bits) +{ + assert(sint_is_in_range(a, num_bits)); + assert(sint_is_in_range(b, num_bits)); + + if (num_bits == 64) { + uint32_t a32[4] = { + (uint32_t)a, + (uint32_t)(a >> 32), + (uint32_t)(a >> 63), /* sign extend */ + (uint32_t)(a >> 63), /* sign extend */ + }; + uint32_t b32[4] = { + (uint32_t)b, + (uint32_t)(b >> 32), + (uint32_t)(b >> 63), /* sign extend */ + (uint32_t)(b >> 63), /* sign extend */ + }; + + uint32_t ab32[4]; + ubm_mul_u32arr(ab32, a32, b32); + + return ab32[2] | ((uint64_t)ab32[3] << 32); + } else { + assert(num_bits <= 32); + return strunc((a * b) >> num_bits, num_bits); + } +} + +static inline uint64_t +fast_udiv_add_sat(uint64_t n, struct util_fast_udiv_info m, unsigned num_bits) +{ + assert(uint_is_in_range(n, num_bits)); + assert(uint_is_in_range(m.multiplier, num_bits)); + + n = n >> m.pre_shift; + n = uadd_sat(n, m.increment, num_bits); + n = umul_add_high(n, m.multiplier, 0, num_bits); + n = n >> m.post_shift; + + return n; +} + +static inline uint64_t +fast_udiv_mul_add(uint64_t n, struct util_fast_udiv_info m, unsigned num_bits) +{ + assert(uint_is_in_range(n, num_bits)); + assert(uint_is_in_range(m.multiplier, num_bits)); + + n = n >> m.pre_shift; + n = umul_add_high(n, m.multiplier, + m.increment ? m.multiplier : 0, + num_bits); + n = n >> m.post_shift; + + return n; +} + +static inline uint64_t +fast_sdiv(int64_t n, int64_t d, struct util_fast_sdiv_info m, unsigned num_bits) +{ + assert(sint_is_in_range(n, num_bits)); + assert(sint_is_in_range(d, num_bits)); + assert(sint_is_in_range(m.multiplier, num_bits)); + + int64_t res; + res = smul_high(n, m.multiplier, num_bits); + if (d > 0 && m.multiplier < 0) + res = strunc(res + n, num_bits); + if (d < 0 && m.multiplier > 0) + res = strunc(res - n, num_bits); + res = res >> m.shift; + res = res - (res >> (num_bits - 1)); + + return res; +} + +static uint64_t +rand_uint(unsigned bits, unsigned min) +{ + assert(bits >= 4); + + /* Make sure we get some small and large numbers and powers of two every + * once in a while + */ + int k = rand() % 64; + if (k == 17) { + return min + (rand() % 16); + } else if (k == 42) { + return MAX_UINT(bits) - (rand() % 16); + } else if (k == 9) { + uint64_t r; + do { + r = 1ull << (rand() % bits); + } while (r < min); + return r; + } + + if (min == 0) { + assert(bits <= 64); + uint64_t r = 0; + for (unsigned i = 0; i < 8; i++) + r |= ((uint64_t)rand() & 0xf) << i * 8; + return r >> (63 - (rand() % bits)); + } else { + uint64_t r; + do { + r = rand_uint(bits, 0); + } while (r < min); + return r; + } +} + +static int64_t +rand_sint(unsigned bits, unsigned min_abs) +{ + /* Make sure we hit MIN_INT every once in a while */ + if (rand() % 64 == 37) + return -(1 << (bits - 1)); + + int64_t s = rand_uint(bits - 1, min_abs); + return rand() & 1 ? s : -s; +} + +static uint64_t +udiv(uint64_t a, uint64_t b, unsigned bit_size) +{ + switch (bit_size) { + case 64: return (uint64_t)a / (uint64_t)b; + case 32: return (uint32_t)a / (uint32_t)b; + case 16: return (uint16_t)a / (uint16_t)b; + case 8: return (uint8_t)a / (uint8_t)b; + default: + assert(!"Invalid bit size"); + return 0; + } +} + +static int64_t +sdiv(int64_t a, int64_t b, unsigned bit_size) +{ + switch (bit_size) { + case 64: return (int64_t)a / (int64_t)b; + case 32: return (int32_t)a / (int32_t)b; + case 16: return (int16_t)a / (int16_t)b; + case 8: return (int8_t)a / (int8_t)b; + default: + assert(!"Invalid bit size"); + return 0; + } +} + +static void +random_udiv_add_sat_test(unsigned bits, bool bounded) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + uint64_t n = rand_uint(bits, 0); + uint64_t d = rand_uint(bits, 2); + assert(uint_is_in_range(n, bits)); + assert(uint_is_in_range(d, bits) && d >= 2); + + unsigned n_bits = bounded ? util_logbase2_64(MAX2(n, 1)) + 1 : bits; + + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, n_bits, bits); + EXPECT_EQ(fast_udiv_add_sat(n, m, bits), udiv(n, d, bits)); + } +} + +static void +random_udiv_mul_add_test(unsigned bits, bool bounded) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + uint64_t n = rand_uint(bits, 0); + uint64_t d = rand_uint(bits, 1); + assert(uint_is_in_range(n, bits)); + assert(uint_is_in_range(d, bits) && d >= 1); + + unsigned n_bits = bounded ? util_logbase2_64(MAX2(n, 1)) + 1: bits; + + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, n_bits, bits); + EXPECT_EQ(fast_udiv_mul_add(n, m, bits), udiv(n, d, bits)); + } +} + +static void +random_sdiv_test(unsigned bits) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + int64_t n = rand_sint(bits, 0); + int64_t d; + do { + d = rand_sint(bits, 2); + } while (util_is_power_of_two_or_zero64(llabs(d))); + + assert(sint_is_in_range(n, bits)); + assert(sint_is_in_range(d, bits) && llabs(d) >= 2); + + struct util_fast_sdiv_info m = + util_compute_fast_sdiv_info(d, bits); + EXPECT_EQ(fast_sdiv(n, d, m, bits), sdiv(n, d, bits)); + } +} + +TEST(fast_idiv_by_const, uint8_add_sat) +{ + /* 8-bit is small enough we can brute-force the entire space */ + for (unsigned d = 2; d < 256; d++) { + for (unsigned n_bits = 1; n_bits <= 8; n_bits++) { + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, n_bits, 8); + + for (unsigned n = 0; n < (1u << n_bits); n++) + EXPECT_EQ(fast_udiv_add_sat(n, m, 8), udiv(n, d, 8)); + } + } +} + +TEST(fast_idiv_by_const, uint8_mul_add) +{ + /* 8-bit is small enough we can brute-force the entire space */ + for (unsigned d = 2; d < 256; d++) { + for (unsigned n_bits = 1; n_bits <= 8; n_bits++) { + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, n_bits, 8); + + for (unsigned n = 0; n < (1u << n_bits); n++) + EXPECT_EQ(fast_udiv_mul_add(n, m, 8), udiv(n, d, 8)); + } + } +} + +TEST(fast_idiv_by_const, int8) +{ + /* 8-bit is small enough we can brute-force the entire space */ + for (int n = -128; n < 128; n++) { + for (int d = -128; d < 128; d++) { + if (util_is_power_of_two_or_zero(abs(d))) + continue; + + struct util_fast_sdiv_info m = + util_compute_fast_sdiv_info(d, 8); + EXPECT_EQ(fast_sdiv(n, d, m, 8), sdiv(n, d, 8)); + } + } +} + +TEST(fast_idiv_by_const, uint16_add_sat_bounded) +{ + random_udiv_add_sat_test(16, true); +} + +TEST(fast_idiv_by_const, uint16_add_sat_full) +{ + random_udiv_add_sat_test(16, false); +} + +TEST(fast_idiv_by_const, uint16_mul_add_bounded) +{ + random_udiv_mul_add_test(16, true); +} + +TEST(fast_idiv_by_const, uint16_mul_add_full) +{ + random_udiv_mul_add_test(16, false); +} + +TEST(fast_idiv_by_const, int16) +{ + random_sdiv_test(16); +} + +TEST(fast_idiv_by_const, uint32_add_sat_bounded) +{ + random_udiv_add_sat_test(32, true); +} + +TEST(fast_idiv_by_const, uint32_add_sat_full) +{ + random_udiv_add_sat_test(32, false); +} + +TEST(fast_idiv_by_const, uint32_mul_add_bounded) +{ + random_udiv_mul_add_test(32, true); +} + +TEST(fast_idiv_by_const, uint32_mul_add_full) +{ + random_udiv_mul_add_test(32, false); +} + +TEST(fast_idiv_by_const, int32) +{ + random_sdiv_test(32); +} + +TEST(fast_idiv_by_const, util_fast_udiv32) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + uint32_t n = rand_uint(32, 0); + uint32_t d = rand_uint(32, 1); + + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, 32, 32); + EXPECT_EQ(util_fast_udiv32(n, m), n / d); + } +} + +TEST(fast_idiv_by_const, util_fast_udiv32_nuw) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + uint32_t n = rand_uint(32, 0); + if (n == UINT32_MAX) + continue; + uint32_t d = rand_uint(32, 1); + + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, 32, 32); + EXPECT_EQ(util_fast_udiv32_nuw(n, m), n / d); + } +} + +TEST(fast_idiv_by_const, util_fast_udiv32_u31_d_not_one) +{ + for (unsigned i = 0; i < RAND_TEST_ITERATIONS; i++) { + uint32_t n = rand_uint(31, 0); + uint32_t d = rand_uint(31, 2); + + struct util_fast_udiv_info m = + util_compute_fast_udiv_info(d, 31, 32); + EXPECT_EQ(util_fast_udiv32_u31_d_not_one(n, m), n / d); + } +} + +TEST(fast_idiv_by_const, uint64_add_sat_bounded) +{ + random_udiv_add_sat_test(64, true); +} + +TEST(fast_idiv_by_const, uint64_add_sat_full) +{ + random_udiv_add_sat_test(64, false); +} + +TEST(fast_idiv_by_const, uint64_mul_add_bounded) +{ + random_udiv_mul_add_test(64, true); +} + +TEST(fast_idiv_by_const, uint64_mul_add_full) +{ + random_udiv_mul_add_test(64, false); +} + +TEST(fast_idiv_by_const, int64) +{ + random_sdiv_test(64); +} diff --git a/src/util/tests/fast_idiv_by_const/meson.build b/src/util/tests/fast_idiv_by_const/meson.build new file mode 100644 index 00000000000..8c3b79493ff --- /dev/null +++ b/src/util/tests/fast_idiv_by_const/meson.build @@ -0,0 +1,30 @@ +# Copyright © 2018 Intel Corporation + +# Permission is hereby granted, free of charge, to any person obtaining a copy +# of this software and associated documentation files (the "Software"), to deal +# in the Software without restriction, including without limitation the rights +# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +# copies of the Software, and to permit persons to whom the Software is +# furnished to do so, subject to the following conditions: + +# The above copyright notice and this permission notice shall be included in +# all copies or substantial portions of the Software. + +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +# SOFTWARE. + +test( + 'fast_idiv_by_const', + executable( + 'fast_idiv_by_const_test', + 'fast_idiv_by_const_test.cpp', + dependencies : [dep_thread, dep_dl, idep_gtest], + include_directories : inc_common, + link_with : [libmesa_util], + ) +) |