1cabdff1aSopenharmony_ci;***************************************************************************** 2cabdff1aSopenharmony_ci;* Assembly testing and benchmarking tool 3cabdff1aSopenharmony_ci;* Copyright (c) 2008 Loren Merritt 4cabdff1aSopenharmony_ci;* Copyright (c) 2012 Henrik Gramner 5cabdff1aSopenharmony_ci;* 6cabdff1aSopenharmony_ci;* This file is part of FFmpeg. 7cabdff1aSopenharmony_ci;* 8cabdff1aSopenharmony_ci;* FFmpeg is free software; you can redistribute it and/or modify 9cabdff1aSopenharmony_ci;* it under the terms of the GNU General Public License as published by 10cabdff1aSopenharmony_ci;* the Free Software Foundation; either version 2 of the License, or 11cabdff1aSopenharmony_ci;* (at your option) any later version. 12cabdff1aSopenharmony_ci;* 13cabdff1aSopenharmony_ci;* FFmpeg is distributed in the hope that it will be useful, 14cabdff1aSopenharmony_ci;* but WITHOUT ANY WARRANTY; without even the implied warranty of 15cabdff1aSopenharmony_ci;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16cabdff1aSopenharmony_ci;* GNU General Public License for more details. 17cabdff1aSopenharmony_ci;* 18cabdff1aSopenharmony_ci;* You should have received a copy of the GNU General Public License 19cabdff1aSopenharmony_ci;* along with this program; if not, write to the Free Software 20cabdff1aSopenharmony_ci;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA. 21cabdff1aSopenharmony_ci;***************************************************************************** 22cabdff1aSopenharmony_ci 23cabdff1aSopenharmony_ci%define private_prefix checkasm 24cabdff1aSopenharmony_ci%include "libavutil/x86/x86inc.asm" 25cabdff1aSopenharmony_ci 26cabdff1aSopenharmony_ciSECTION_RODATA 27cabdff1aSopenharmony_ci 28cabdff1aSopenharmony_cierror_message: db "failed to preserve register", 0 29cabdff1aSopenharmony_cierror_message_emms: db "failed to issue emms", 0 30cabdff1aSopenharmony_ci 31cabdff1aSopenharmony_ci%if ARCH_X86_64 32cabdff1aSopenharmony_ci; just random numbers to reduce the chance of incidental match 33cabdff1aSopenharmony_ciALIGN 16 34cabdff1aSopenharmony_cix6: dq 0x1a1b2550a612b48c,0x79445c159ce79064 35cabdff1aSopenharmony_cix7: dq 0x2eed899d5a28ddcd,0x86b2536fcd8cf636 36cabdff1aSopenharmony_cix8: dq 0xb0856806085e7943,0x3f2bf84fc0fcca4e 37cabdff1aSopenharmony_cix9: dq 0xacbd382dcf5b8de2,0xd229e1f5b281303f 38cabdff1aSopenharmony_cix10: dq 0x71aeaff20b095fd9,0xab63e2e11fa38ed9 39cabdff1aSopenharmony_cix11: dq 0x89b0c0765892729a,0x77d410d5c42c882d 40cabdff1aSopenharmony_cix12: dq 0xc45ea11a955d8dd5,0x24b3c1d2a024048b 41cabdff1aSopenharmony_cix13: dq 0x2e8ec680de14b47c,0xdd7b8919edd42786 42cabdff1aSopenharmony_cix14: dq 0x135ce6888fa02cbf,0x11e53e2b2ac655ef 43cabdff1aSopenharmony_cix15: dq 0x011ff554472a7a10,0x6de8f4c914c334d5 44cabdff1aSopenharmony_cin7: dq 0x21f86d66c8ca00ce 45cabdff1aSopenharmony_cin8: dq 0x75b6ba21077c48ad 46cabdff1aSopenharmony_cin9: dq 0xed56bb2dcb3c7736 47cabdff1aSopenharmony_cin10: dq 0x8bda43d3fd1a7e06 48cabdff1aSopenharmony_cin11: dq 0xb64a9c9e5d318408 49cabdff1aSopenharmony_cin12: dq 0xdf9a54b303f1d3a3 50cabdff1aSopenharmony_cin13: dq 0x4a75479abd64e097 51cabdff1aSopenharmony_cin14: dq 0x249214109d5d1c88 52cabdff1aSopenharmony_ci%endif 53cabdff1aSopenharmony_ci 54cabdff1aSopenharmony_ciSECTION .text 55cabdff1aSopenharmony_ci 56cabdff1aSopenharmony_cicextern fail_func 57cabdff1aSopenharmony_ci 58cabdff1aSopenharmony_ci; max number of args used by any asm function. 59cabdff1aSopenharmony_ci; (max_args % 4) must equal 3 for stack alignment 60cabdff1aSopenharmony_ci%define max_args 15 61cabdff1aSopenharmony_ci 62cabdff1aSopenharmony_ci%if ARCH_X86_64 63cabdff1aSopenharmony_ci 64cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 65cabdff1aSopenharmony_ci; int checkasm_stack_clobber(uint64_t clobber, ...) 66cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 67cabdff1aSopenharmony_cicglobal stack_clobber, 1,2 68cabdff1aSopenharmony_ci ; Clobber the stack with junk below the stack pointer 69cabdff1aSopenharmony_ci %define argsize (max_args+6)*8 70cabdff1aSopenharmony_ci SUB rsp, argsize 71cabdff1aSopenharmony_ci mov r1, argsize-8 72cabdff1aSopenharmony_ci.loop: 73cabdff1aSopenharmony_ci mov [rsp+r1], r0 74cabdff1aSopenharmony_ci sub r1, 8 75cabdff1aSopenharmony_ci jge .loop 76cabdff1aSopenharmony_ci ADD rsp, argsize 77cabdff1aSopenharmony_ci RET 78cabdff1aSopenharmony_ci 79cabdff1aSopenharmony_ci%if WIN64 80cabdff1aSopenharmony_ci %assign free_regs 7 81cabdff1aSopenharmony_ci DECLARE_REG_TMP 4 82cabdff1aSopenharmony_ci%else 83cabdff1aSopenharmony_ci %assign free_regs 9 84cabdff1aSopenharmony_ci DECLARE_REG_TMP 7 85cabdff1aSopenharmony_ci%endif 86cabdff1aSopenharmony_ci 87cabdff1aSopenharmony_ci%macro report_fail 1 88cabdff1aSopenharmony_ci mov r9, rax 89cabdff1aSopenharmony_ci mov r10, rdx 90cabdff1aSopenharmony_ci lea r0, [%1] 91cabdff1aSopenharmony_ci xor eax, eax 92cabdff1aSopenharmony_ci call fail_func 93cabdff1aSopenharmony_ci mov rdx, r10 94cabdff1aSopenharmony_ci mov rax, r9 95cabdff1aSopenharmony_ci%endmacro 96cabdff1aSopenharmony_ci 97cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 98cabdff1aSopenharmony_ci; void checkasm_checked_call(void *func, ...) 99cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 100cabdff1aSopenharmony_ciINIT_XMM 101cabdff1aSopenharmony_ci%macro CHECKED_CALL 0-1 102cabdff1aSopenharmony_cicglobal checked_call%1, 2,15,16,max_args*8+8 103cabdff1aSopenharmony_ci mov t0, r0 104cabdff1aSopenharmony_ci 105cabdff1aSopenharmony_ci ; All arguments have been pushed on the stack instead of registers in order to 106cabdff1aSopenharmony_ci ; test for incorrect assumptions that 32-bit ints are zero-extended to 64-bit. 107cabdff1aSopenharmony_ci mov r0, r6mp 108cabdff1aSopenharmony_ci mov r1, r7mp 109cabdff1aSopenharmony_ci mov r2, r8mp 110cabdff1aSopenharmony_ci mov r3, r9mp 111cabdff1aSopenharmony_ci%if UNIX64 112cabdff1aSopenharmony_ci mov r4, r10mp 113cabdff1aSopenharmony_ci mov r5, r11mp 114cabdff1aSopenharmony_ci %assign i 6 115cabdff1aSopenharmony_ci %rep max_args-6 116cabdff1aSopenharmony_ci mov r9, [rsp+stack_offset+(i+1)*8] 117cabdff1aSopenharmony_ci mov [rsp+(i-6)*8], r9 118cabdff1aSopenharmony_ci %assign i i+1 119cabdff1aSopenharmony_ci %endrep 120cabdff1aSopenharmony_ci%else ; WIN64 121cabdff1aSopenharmony_ci %assign i 4 122cabdff1aSopenharmony_ci %rep max_args-4 123cabdff1aSopenharmony_ci mov r9, [rsp+stack_offset+(i+7)*8] 124cabdff1aSopenharmony_ci mov [rsp+i*8], r9 125cabdff1aSopenharmony_ci %assign i i+1 126cabdff1aSopenharmony_ci %endrep 127cabdff1aSopenharmony_ci 128cabdff1aSopenharmony_ci ; Move possible floating-point arguments to the correct registers 129cabdff1aSopenharmony_ci movq m0, r0 130cabdff1aSopenharmony_ci movq m1, r1 131cabdff1aSopenharmony_ci movq m2, r2 132cabdff1aSopenharmony_ci movq m3, r3 133cabdff1aSopenharmony_ci 134cabdff1aSopenharmony_ci %assign i 6 135cabdff1aSopenharmony_ci %rep 16-6 136cabdff1aSopenharmony_ci mova m %+ i, [x %+ i] 137cabdff1aSopenharmony_ci %assign i i+1 138cabdff1aSopenharmony_ci %endrep 139cabdff1aSopenharmony_ci%endif 140cabdff1aSopenharmony_ci 141cabdff1aSopenharmony_ci%assign i 14 142cabdff1aSopenharmony_ci%rep 15-free_regs 143cabdff1aSopenharmony_ci mov r %+ i, [n %+ i] 144cabdff1aSopenharmony_ci %assign i i-1 145cabdff1aSopenharmony_ci%endrep 146cabdff1aSopenharmony_ci call t0 147cabdff1aSopenharmony_ci%assign i 14 148cabdff1aSopenharmony_ci%rep 15-free_regs 149cabdff1aSopenharmony_ci xor r %+ i, [n %+ i] 150cabdff1aSopenharmony_ci or r14, r %+ i 151cabdff1aSopenharmony_ci %assign i i-1 152cabdff1aSopenharmony_ci%endrep 153cabdff1aSopenharmony_ci 154cabdff1aSopenharmony_ci%if WIN64 155cabdff1aSopenharmony_ci %assign i 6 156cabdff1aSopenharmony_ci %rep 16-6 157cabdff1aSopenharmony_ci pxor m %+ i, [x %+ i] 158cabdff1aSopenharmony_ci por m6, m %+ i 159cabdff1aSopenharmony_ci %assign i i+1 160cabdff1aSopenharmony_ci %endrep 161cabdff1aSopenharmony_ci packsswb m6, m6 162cabdff1aSopenharmony_ci movq r5, m6 163cabdff1aSopenharmony_ci or r14, r5 164cabdff1aSopenharmony_ci%endif 165cabdff1aSopenharmony_ci 166cabdff1aSopenharmony_ci ; Call fail_func() with a descriptive message to mark it as a failure 167cabdff1aSopenharmony_ci ; if the called function didn't preserve all callee-saved registers. 168cabdff1aSopenharmony_ci ; Save the return value located in rdx:rax first to prevent clobbering. 169cabdff1aSopenharmony_ci jz .clobber_ok 170cabdff1aSopenharmony_ci report_fail error_message 171cabdff1aSopenharmony_ci.clobber_ok: 172cabdff1aSopenharmony_ci%ifidn %1, _emms 173cabdff1aSopenharmony_ci emms 174cabdff1aSopenharmony_ci%elifnidn %1, _float 175cabdff1aSopenharmony_ci fstenv [rsp] 176cabdff1aSopenharmony_ci cmp word [rsp + 8], 0xffff 177cabdff1aSopenharmony_ci je .emms_ok 178cabdff1aSopenharmony_ci report_fail error_message_emms 179cabdff1aSopenharmony_ci emms 180cabdff1aSopenharmony_ci.emms_ok: 181cabdff1aSopenharmony_ci%endif 182cabdff1aSopenharmony_ci RET 183cabdff1aSopenharmony_ci%endmacro 184cabdff1aSopenharmony_ci 185cabdff1aSopenharmony_ci%else 186cabdff1aSopenharmony_ci 187cabdff1aSopenharmony_ci; just random numbers to reduce the chance of incidental match 188cabdff1aSopenharmony_ci%define n3 dword 0x6549315c 189cabdff1aSopenharmony_ci%define n4 dword 0xe02f3e23 190cabdff1aSopenharmony_ci%define n5 dword 0xb78d0d1d 191cabdff1aSopenharmony_ci%define n6 dword 0x33627ba7 192cabdff1aSopenharmony_ci 193cabdff1aSopenharmony_ci%macro report_fail 1 194cabdff1aSopenharmony_ci mov r3, eax 195cabdff1aSopenharmony_ci mov r4, edx 196cabdff1aSopenharmony_ci lea r0, [%1] 197cabdff1aSopenharmony_ci mov [esp], r0 198cabdff1aSopenharmony_ci call fail_func 199cabdff1aSopenharmony_ci mov edx, r4 200cabdff1aSopenharmony_ci mov eax, r3 201cabdff1aSopenharmony_ci%endmacro 202cabdff1aSopenharmony_ci 203cabdff1aSopenharmony_ci%macro CHECKED_CALL 0-1 204cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 205cabdff1aSopenharmony_ci; void checkasm_checked_call(void *func, ...) 206cabdff1aSopenharmony_ci;----------------------------------------------------------------------------- 207cabdff1aSopenharmony_cicglobal checked_call%1, 1,7 208cabdff1aSopenharmony_ci mov r3, n3 209cabdff1aSopenharmony_ci mov r4, n4 210cabdff1aSopenharmony_ci mov r5, n5 211cabdff1aSopenharmony_ci mov r6, n6 212cabdff1aSopenharmony_ci%rep max_args 213cabdff1aSopenharmony_ci PUSH dword [esp+20+max_args*4] 214cabdff1aSopenharmony_ci%endrep 215cabdff1aSopenharmony_ci call r0 216cabdff1aSopenharmony_ci xor r3, n3 217cabdff1aSopenharmony_ci xor r4, n4 218cabdff1aSopenharmony_ci xor r5, n5 219cabdff1aSopenharmony_ci xor r6, n6 220cabdff1aSopenharmony_ci or r3, r4 221cabdff1aSopenharmony_ci or r5, r6 222cabdff1aSopenharmony_ci or r3, r5 223cabdff1aSopenharmony_ci jz .clobber_ok 224cabdff1aSopenharmony_ci report_fail error_message 225cabdff1aSopenharmony_ci.clobber_ok: 226cabdff1aSopenharmony_ci%ifidn %1, _emms 227cabdff1aSopenharmony_ci emms 228cabdff1aSopenharmony_ci%elifnidn %1, _float 229cabdff1aSopenharmony_ci fstenv [esp] 230cabdff1aSopenharmony_ci cmp word [esp + 8], 0xffff 231cabdff1aSopenharmony_ci je .emms_ok 232cabdff1aSopenharmony_ci report_fail error_message_emms 233cabdff1aSopenharmony_ci emms 234cabdff1aSopenharmony_ci.emms_ok: 235cabdff1aSopenharmony_ci%endif 236cabdff1aSopenharmony_ci add esp, max_args*4 237cabdff1aSopenharmony_ci REP_RET 238cabdff1aSopenharmony_ci%endmacro 239cabdff1aSopenharmony_ci 240cabdff1aSopenharmony_ci%endif ; ARCH_X86_64 241cabdff1aSopenharmony_ci 242cabdff1aSopenharmony_ciCHECKED_CALL 243cabdff1aSopenharmony_ciCHECKED_CALL _emms 244cabdff1aSopenharmony_ciCHECKED_CALL _float 245