| 1 | /***************************************************************************** |
| 2 | * Copyright (C) 2013 x265 project |
| 3 | * |
| 4 | * Authors: Steve Borho <steve@borho.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. |
| 19 | * |
| 20 | * This program is also available under a commercial proprietary license. |
| 21 | * For more information, contact us at license @ x265.com. |
| 22 | *****************************************************************************/ |
| 23 | |
| 24 | #ifndef _TESTHARNESS_H_ |
| 25 | #define _TESTHARNESS_H_ 1 |
| 26 | |
| 27 | #include "common.h" |
| 28 | #include "primitives.h" |
| 29 | |
| 30 | #if _MSC_VER |
| 31 | #pragma warning(disable: 4324) // structure was padded due to __declspec(align()) |
| 32 | #endif |
| 33 | |
| 34 | #if HIGH_BIT_DEPTH |
| 35 | #define BIT_DEPTH 10 |
| 36 | #else |
| 37 | #define BIT_DEPTH 8 |
| 38 | #endif |
| 39 | #define PIXEL_MAX ((1 << BIT_DEPTH) - 1) |
| 40 | #define PIXEL_MIN 0 |
| 41 | #define SHORT_MAX 32767 |
| 42 | #define SHORT_MIN -32767 |
| 43 | #define UNSIGNED_SHORT_MAX 65535 |
| 44 | |
| 45 | using namespace x265; |
| 46 | |
| 47 | extern const char* lumaPartStr[NUM_LUMA_PARTITIONS]; |
| 48 | extern const char* const* chromaPartStr[X265_CSP_COUNT]; |
| 49 | |
| 50 | class TestHarness |
| 51 | { |
| 52 | public: |
| 53 | |
| 54 | TestHarness() {} |
| 55 | |
| 56 | virtual ~TestHarness() {} |
| 57 | |
| 58 | virtual bool testCorrectness(const EncoderPrimitives& ref, const EncoderPrimitives& opt) = 0; |
| 59 | |
| 60 | virtual void measureSpeed(const EncoderPrimitives& ref, const EncoderPrimitives& opt) = 0; |
| 61 | |
| 62 | virtual const char *getName() const = 0; |
| 63 | |
| 64 | protected: |
| 65 | |
| 66 | /* Temporary variables for stack checks */ |
| 67 | int m_ok; |
| 68 | |
| 69 | uint64_t m_rand; |
| 70 | }; |
| 71 | |
| 72 | #ifdef _MSC_VER |
| 73 | #include <intrin.h> |
| 74 | #elif HAVE_RDTSC |
| 75 | #include <intrin.h> |
| 76 | #elif defined(__GNUC__) |
| 77 | /* fallback for older GCC/MinGW */ |
| 78 | static inline uint32_t __rdtsc(void) |
| 79 | { |
| 80 | uint32_t a = 0; |
| 81 | |
| 82 | asm volatile("rdtsc" : "=a" (a) ::"edx"); |
| 83 | return a; |
| 84 | } |
| 85 | |
| 86 | #endif // ifdef _MSC_VER |
| 87 | |
| 88 | #define BENCH_RUNS 1000 |
| 89 | |
| 90 | // Adapted from checkasm.c, runs each optimized primitive four times, measures rdtsc |
| 91 | // and discards invalid times. Repeats 1000 times to get a good average. Then measures |
| 92 | // the C reference with fewer runs and reports X factor and average cycles. |
| 93 | #define REPORT_SPEEDUP(RUNOPT, RUNREF, ...) \ |
| 94 | { \ |
| 95 | uint32_t cycles = 0; int runs = 0; \ |
| 96 | RUNOPT(__VA_ARGS__); \ |
| 97 | for (int ti = 0; ti < BENCH_RUNS; ti++) { \ |
| 98 | uint32_t t0 = (uint32_t)__rdtsc(); \ |
| 99 | RUNOPT(__VA_ARGS__); \ |
| 100 | RUNOPT(__VA_ARGS__); \ |
| 101 | RUNOPT(__VA_ARGS__); \ |
| 102 | RUNOPT(__VA_ARGS__); \ |
| 103 | uint32_t t1 = (uint32_t)__rdtsc() - t0; \ |
| 104 | if (t1 * runs <= cycles * 4 && ti > 0) { cycles += t1; runs++; } \ |
| 105 | } \ |
| 106 | uint32_t refcycles = 0; int refruns = 0; \ |
| 107 | RUNREF(__VA_ARGS__); \ |
| 108 | for (int ti = 0; ti < BENCH_RUNS / 4; ti++) { \ |
| 109 | uint32_t t0 = (uint32_t)__rdtsc(); \ |
| 110 | RUNREF(__VA_ARGS__); \ |
| 111 | RUNREF(__VA_ARGS__); \ |
| 112 | RUNREF(__VA_ARGS__); \ |
| 113 | RUNREF(__VA_ARGS__); \ |
| 114 | uint32_t t1 = (uint32_t)__rdtsc() - t0; \ |
| 115 | if (t1 * refruns <= refcycles * 4 && ti > 0) { refcycles += t1; refruns++; } \ |
| 116 | } \ |
| 117 | x265_emms(); \ |
| 118 | float optperf = (10.0f * cycles / runs) / 4; \ |
| 119 | float refperf = (10.0f * refcycles / refruns) / 4; \ |
| 120 | printf("\t%3.2fx ", refperf / optperf); \ |
| 121 | printf("\t %-8.2lf \t %-8.2lf\n", optperf, refperf); \ |
| 122 | } |
| 123 | |
| 124 | extern "C" { |
| 125 | #if X265_ARCH_X86 |
| 126 | int x265_stack_pagealign(int (*func)(), int align); |
| 127 | |
| 128 | /* detect when callee-saved regs aren't saved |
| 129 | * needs an explicit asm check because it only sometimes crashes in normal use. */ |
| 130 | intptr_t x265_checkasm_call(intptr_t (*func)(), int *ok, ...); |
| 131 | float x265_checkasm_call_float(float (*func)(), int *ok, ...); |
| 132 | #else |
| 133 | #define x265_stack_pagealign(func, align) func() |
| 134 | #endif |
| 135 | |
| 136 | #if X86_64 |
| 137 | |
| 138 | /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit. |
| 139 | * This is done by clobbering the stack with junk around the stack pointer and calling the |
| 140 | * assembly function through x265_checkasm_call with added dummy arguments which forces all |
| 141 | * real arguments to be passed on the stack and not in registers. For 32-bit argument the |
| 142 | * upper half of the 64-bit register location on the stack will now contain junk. Note that |
| 143 | * this is dependent on compiler behavior and that interrupts etc. at the wrong time may |
| 144 | * overwrite the junk written to the stack so there's no guarantee that it will always |
| 145 | * detect all functions that assumes zero-extension. |
| 146 | */ |
| 147 | void x265_checkasm_stack_clobber(uint64_t clobber, ...); |
| 148 | #define checked(func, ...) ( \ |
| 149 | m_ok = 1, m_rand = (rand() & 0xffff) * 0x0001000100010001ULL, \ |
| 150 | x265_checkasm_stack_clobber(m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, \ |
| 151 | m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, \ |
| 152 | m_rand, m_rand, m_rand, m_rand, m_rand), /* max_args+6 */ \ |
| 153 | x265_checkasm_call((intptr_t(*)())func, &m_ok, 0, 0, 0, 0, __VA_ARGS__)) |
| 154 | |
| 155 | #define checked_float(func, ...) ( \ |
| 156 | m_ok = 1, m_rand = (rand() & 0xffff) * 0x0001000100010001ULL, \ |
| 157 | x265_checkasm_stack_clobber(m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, \ |
| 158 | m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, m_rand, \ |
| 159 | m_rand, m_rand, m_rand, m_rand, m_rand), /* max_args+6 */ \ |
| 160 | x265_checkasm_call_float((float(*)())func, &m_ok, 0, 0, 0, 0, __VA_ARGS__)) |
| 161 | #define reportfail() if (!m_ok) { fprintf(stderr, "stack clobber check failed at %s:%d", __FILE__, __LINE__); abort(); } |
| 162 | #elif ARCH_X86 |
| 163 | #define checked(func, ...) x265_checkasm_call((intptr_t(*)())func, &m_ok, __VA_ARGS__); |
| 164 | #define checked_float(func, ...) x265_checkasm_call_float((float(*)())func, &m_ok, __VA_ARGS__); |
| 165 | |
| 166 | #else // if X86_64 |
| 167 | #define checked(func, ...) func(__VA_ARGS__) |
| 168 | #define checked_float(func, ...) func(__VA_ARGS__) |
| 169 | #define reportfail() |
| 170 | #endif // if X86_64 |
| 171 | } |
| 172 | |
| 173 | #endif // ifndef _TESTHARNESS_H_ |