1; 2; Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3; 4; Use of this source code is governed by a BSD-style license 5; that can be found in the LICENSE file in the root of the source 6; tree. An additional intellectual property rights grant can be found 7; in the file PATENTS. All contributing project authors may 8; be found in the AUTHORS file in the root of the source tree. 9; 10 11 12%include "vpx_config.asm" 13 14; 32/64 bit compatibility macros 15; 16; In general, we make the source use 64 bit syntax, then twiddle with it using 17; the preprocessor to get the 32 bit syntax on 32 bit platforms. 18; 19%ifidn __OUTPUT_FORMAT__,elf32 20%define ABI_IS_32BIT 1 21%elifidn __OUTPUT_FORMAT__,macho32 22%define ABI_IS_32BIT 1 23%elifidn __OUTPUT_FORMAT__,win32 24%define ABI_IS_32BIT 1 25%elifidn __OUTPUT_FORMAT__,aout 26%define ABI_IS_32BIT 1 27%else 28%define ABI_IS_32BIT 0 29%endif 30 31%if ABI_IS_32BIT 32%define rax eax 33%define rbx ebx 34%define rcx ecx 35%define rdx edx 36%define rsi esi 37%define rdi edi 38%define rsp esp 39%define rbp ebp 40%define movsxd mov 41%macro movq 2 42 %ifidn %1,eax 43 movd %1,%2 44 %elifidn %2,eax 45 movd %1,%2 46 %elifidn %1,ebx 47 movd %1,%2 48 %elifidn %2,ebx 49 movd %1,%2 50 %elifidn %1,ecx 51 movd %1,%2 52 %elifidn %2,ecx 53 movd %1,%2 54 %elifidn %1,edx 55 movd %1,%2 56 %elifidn %2,edx 57 movd %1,%2 58 %elifidn %1,esi 59 movd %1,%2 60 %elifidn %2,esi 61 movd %1,%2 62 %elifidn %1,edi 63 movd %1,%2 64 %elifidn %2,edi 65 movd %1,%2 66 %elifidn %1,esp 67 movd %1,%2 68 %elifidn %2,esp 69 movd %1,%2 70 %elifidn %1,ebp 71 movd %1,%2 72 %elifidn %2,ebp 73 movd %1,%2 74 %else 75 movq %1,%2 76 %endif 77%endmacro 78%endif 79 80 81; LIBVPX_YASM_WIN64 82; Set LIBVPX_YASM_WIN64 if output is Windows 64bit so the code will work if x64 83; or win64 is defined on the Yasm command line. 84%ifidn __OUTPUT_FORMAT__,win64 85%define LIBVPX_YASM_WIN64 1 86%elifidn __OUTPUT_FORMAT__,x64 87%define LIBVPX_YASM_WIN64 1 88%else 89%define LIBVPX_YASM_WIN64 0 90%endif 91 92; sym() 93; Return the proper symbol name for the target ABI. 94; 95; Certain ABIs, notably MS COFF and Darwin MACH-O, require that symbols 96; with C linkage be prefixed with an underscore. 97; 98%ifidn __OUTPUT_FORMAT__,elf32 99%define sym(x) x 100%elifidn __OUTPUT_FORMAT__,elf64 101%define sym(x) x 102%elifidn __OUTPUT_FORMAT__,elfx32 103%define sym(x) x 104%elif LIBVPX_YASM_WIN64 105%define sym(x) x 106%else 107%define sym(x) _ %+ x 108%endif 109 110; PRIVATE 111; Macro for the attribute to hide a global symbol for the target ABI. 112; This is only active if CHROMIUM is defined. 113; 114; Chromium doesn't like exported global symbols due to symbol clashing with 115; plugins among other things. 116; 117; Requires Chromium's patched copy of yasm: 118; http://src.chromium.org/viewvc/chrome?view=rev&revision=73761 119; http://www.tortall.net/projects/yasm/ticket/236 120; 121%ifdef CHROMIUM 122 %ifidn __OUTPUT_FORMAT__,elf32 123 %define PRIVATE :hidden 124 %elifidn __OUTPUT_FORMAT__,elf64 125 %define PRIVATE :hidden 126 %elifidn __OUTPUT_FORMAT__,elfx32 127 %define PRIVATE :hidden 128 %elif LIBVPX_YASM_WIN64 129 %define PRIVATE 130 %else 131 %define PRIVATE :private_extern 132 %endif 133%else 134 %define PRIVATE 135%endif 136 137; arg() 138; Return the address specification of the given argument 139; 140%if ABI_IS_32BIT 141 %define arg(x) [ebp+8+4*x] 142%else 143 ; 64 bit ABI passes arguments in registers. This is a workaround to get up 144 ; and running quickly. Relies on SHADOW_ARGS_TO_STACK 145 %if LIBVPX_YASM_WIN64 146 %define arg(x) [rbp+16+8*x] 147 %else 148 %define arg(x) [rbp-8-8*x] 149 %endif 150%endif 151 152; REG_SZ_BYTES, REG_SZ_BITS 153; Size of a register 154%if ABI_IS_32BIT 155%define REG_SZ_BYTES 4 156%define REG_SZ_BITS 32 157%else 158%define REG_SZ_BYTES 8 159%define REG_SZ_BITS 64 160%endif 161 162 163; ALIGN_STACK <alignment> <register> 164; This macro aligns the stack to the given alignment (in bytes). The stack 165; is left such that the previous value of the stack pointer is the first 166; argument on the stack (ie, the inverse of this macro is 'pop rsp.') 167; This macro uses one temporary register, which is not preserved, and thus 168; must be specified as an argument. 169%macro ALIGN_STACK 2 170 mov %2, rsp 171 and rsp, -%1 172 lea rsp, [rsp - (%1 - REG_SZ_BYTES)] 173 push %2 174%endmacro 175 176 177; 178; The Microsoft assembler tries to impose a certain amount of type safety in 179; its register usage. YASM doesn't recognize these directives, so we just 180; %define them away to maintain as much compatibility as possible with the 181; original inline assembler we're porting from. 182; 183%idefine PTR 184%idefine XMMWORD 185%idefine MMWORD 186 187; PIC macros 188; 189%if ABI_IS_32BIT 190 %if CONFIG_PIC=1 191 %ifidn __OUTPUT_FORMAT__,elf32 192 %define WRT_PLT wrt ..plt 193 %macro GET_GOT 1 194 extern _GLOBAL_OFFSET_TABLE_ 195 push %1 196 call %%get_got 197 %%sub_offset: 198 jmp %%exitGG 199 %%get_got: 200 mov %1, [esp] 201 add %1, _GLOBAL_OFFSET_TABLE_ + $$ - %%sub_offset wrt ..gotpc 202 ret 203 %%exitGG: 204 %undef GLOBAL 205 %define GLOBAL(x) x + %1 wrt ..gotoff 206 %undef RESTORE_GOT 207 %define RESTORE_GOT pop %1 208 %endmacro 209 %elifidn __OUTPUT_FORMAT__,macho32 210 %macro GET_GOT 1 211 push %1 212 call %%get_got 213 %%get_got: 214 pop %1 215 %undef GLOBAL 216 %define GLOBAL(x) x + %1 - %%get_got 217 %undef RESTORE_GOT 218 %define RESTORE_GOT pop %1 219 %endmacro 220 %endif 221 %endif 222 223 %ifdef CHROMIUM 224 %ifidn __OUTPUT_FORMAT__,macho32 225 %define HIDDEN_DATA(x) x:private_extern 226 %else 227 %define HIDDEN_DATA(x) x 228 %endif 229 %else 230 %define HIDDEN_DATA(x) x 231 %endif 232%else 233 %macro GET_GOT 1 234 %endmacro 235 %define GLOBAL(x) rel x 236 %ifidn __OUTPUT_FORMAT__,elf64 237 %define WRT_PLT wrt ..plt 238 %define HIDDEN_DATA(x) x:data hidden 239 %elifidn __OUTPUT_FORMAT__,elfx32 240 %define WRT_PLT wrt ..plt 241 %define HIDDEN_DATA(x) x:data hidden 242 %elifidn __OUTPUT_FORMAT__,macho64 243 %ifdef CHROMIUM 244 %define HIDDEN_DATA(x) x:private_extern 245 %else 246 %define HIDDEN_DATA(x) x 247 %endif 248 %else 249 %define HIDDEN_DATA(x) x 250 %endif 251%endif 252%ifnmacro GET_GOT 253 %macro GET_GOT 1 254 %endmacro 255 %define GLOBAL(x) x 256%endif 257%ifndef RESTORE_GOT 258%define RESTORE_GOT 259%endif 260%ifndef WRT_PLT 261%define WRT_PLT 262%endif 263 264%if ABI_IS_32BIT 265 %macro SHADOW_ARGS_TO_STACK 1 266 %endm 267 %define UNSHADOW_ARGS 268%else 269%if LIBVPX_YASM_WIN64 270 %macro SHADOW_ARGS_TO_STACK 1 ; argc 271 %if %1 > 0 272 mov arg(0),rcx 273 %endif 274 %if %1 > 1 275 mov arg(1),rdx 276 %endif 277 %if %1 > 2 278 mov arg(2),r8 279 %endif 280 %if %1 > 3 281 mov arg(3),r9 282 %endif 283 %endm 284%else 285 %macro SHADOW_ARGS_TO_STACK 1 ; argc 286 %if %1 > 0 287 push rdi 288 %endif 289 %if %1 > 1 290 push rsi 291 %endif 292 %if %1 > 2 293 push rdx 294 %endif 295 %if %1 > 3 296 push rcx 297 %endif 298 %if %1 > 4 299 push r8 300 %endif 301 %if %1 > 5 302 push r9 303 %endif 304 %if %1 > 6 305 %assign i %1-6 306 %assign off 16 307 %rep i 308 mov rax,[rbp+off] 309 push rax 310 %assign off off+8 311 %endrep 312 %endif 313 %endm 314%endif 315 %define UNSHADOW_ARGS mov rsp, rbp 316%endif 317 318; Win64 ABI requires that XMM6:XMM15 are callee saved 319; SAVE_XMM n, [u] 320; store registers 6-n on the stack 321; if u is specified, use unaligned movs. 322; Win64 ABI requires 16 byte stack alignment, but then pushes an 8 byte return 323; value. Typically we follow this up with 'push rbp' - re-aligning the stack - 324; but in some cases this is not done and unaligned movs must be used. 325%if LIBVPX_YASM_WIN64 326%macro SAVE_XMM 1-2 a 327 %if %1 < 6 328 %error Only xmm registers 6-15 must be preserved 329 %else 330 %assign last_xmm %1 331 %define movxmm movdq %+ %2 332 %assign xmm_stack_space ((last_xmm - 5) * 16) 333 sub rsp, xmm_stack_space 334 %assign i 6 335 %rep (last_xmm - 5) 336 movxmm [rsp + ((i - 6) * 16)], xmm %+ i 337 %assign i i+1 338 %endrep 339 %endif 340%endmacro 341%macro RESTORE_XMM 0 342 %ifndef last_xmm 343 %error RESTORE_XMM must be paired with SAVE_XMM n 344 %else 345 %assign i last_xmm 346 %rep (last_xmm - 5) 347 movxmm xmm %+ i, [rsp +((i - 6) * 16)] 348 %assign i i-1 349 %endrep 350 add rsp, xmm_stack_space 351 ; there are a couple functions which return from multiple places. 352 ; otherwise, we could uncomment these: 353 ; %undef last_xmm 354 ; %undef xmm_stack_space 355 ; %undef movxmm 356 %endif 357%endmacro 358%else 359%macro SAVE_XMM 1-2 360%endmacro 361%macro RESTORE_XMM 0 362%endmacro 363%endif 364 365; Name of the rodata section 366; 367; .rodata seems to be an elf-ism, as it doesn't work on OSX. 368; 369%ifidn __OUTPUT_FORMAT__,macho64 370%define SECTION_RODATA section .text 371%elifidn __OUTPUT_FORMAT__,macho32 372%macro SECTION_RODATA 0 373section .text 374%endmacro 375%elifidn __OUTPUT_FORMAT__,aout 376%define SECTION_RODATA section .data 377%else 378%define SECTION_RODATA section .rodata 379%endif 380 381 382; Tell GNU ld that we don't require an executable stack. 383%ifidn __OUTPUT_FORMAT__,elf32 384section .note.GNU-stack noalloc noexec nowrite progbits 385section .text 386%elifidn __OUTPUT_FORMAT__,elf64 387section .note.GNU-stack noalloc noexec nowrite progbits 388section .text 389%elifidn __OUTPUT_FORMAT__,elfx32 390section .note.GNU-stack noalloc noexec nowrite progbits 391section .text 392%endif 393 394; On Android platforms use lrand48 when building postproc routines. Prior to L 395; rand() was not available. 396%if CONFIG_POSTPROC=1 || CONFIG_VP9_POSTPROC=1 397%ifdef __ANDROID__ 398extern sym(lrand48) 399%define LIBVPX_RAND lrand48 400%else 401extern sym(rand) 402%define LIBVPX_RAND rand 403%endif 404%endif ; CONFIG_POSTPROC || CONFIG_VP9_POSTPROC 405