;*****************************************************************************
;* checkasm-a.asm: assembly check tool
;*****************************************************************************
-;* Copyright (C) 2008-2012 x264 project
+;* Copyright (C) 2008-2015 x264 project
;*
;* Authors: Loren Merritt <lorenm@u.washington.edu>
+;* Henrik Gramner <henrik@gramner.com>
;*
;* This program is free software; you can redistribute it and/or modify
;* it under the terms of the GNU General Public License as published by
error_message: db "failed to preserve register", 0
-%if WIN64
+%if ARCH_X86_64
; just random numbers to reduce the chance of incidental match
ALIGN 16
-x6: ddq 0x79445c159ce790641a1b2550a612b48c
-x7: ddq 0x86b2536fcd8cf6362eed899d5a28ddcd
-x8: ddq 0x3f2bf84fc0fcca4eb0856806085e7943
-x9: ddq 0xd229e1f5b281303facbd382dcf5b8de2
-x10: ddq 0xab63e2e11fa38ed971aeaff20b095fd9
-x11: ddq 0x77d410d5c42c882d89b0c0765892729a
-x12: ddq 0x24b3c1d2a024048bc45ea11a955d8dd5
-x13: ddq 0xdd7b8919edd427862e8ec680de14b47c
-x14: ddq 0x11e53e2b2ac655ef135ce6888fa02cbf
-x15: ddq 0x6de8f4c914c334d5011ff554472a7a10
-n7: dq 0x21f86d66c8ca00ce
-n8: dq 0x75b6ba21077c48ad
-n9: dq 0xed56bb2dcb3c7736
-n10: dq 0x8bda43d3fd1a7e06
-n11: dq 0xb64a9c9e5d318408
-n12: dq 0xdf9a54b303f1d3a3
-n13: dq 0x4a75479abd64e097
-n14: dq 0x249214109d5d1c88
+x6: dq 0x1a1b2550a612b48c,0x79445c159ce79064
+x7: dq 0x2eed899d5a28ddcd,0x86b2536fcd8cf636
+x8: dq 0xb0856806085e7943,0x3f2bf84fc0fcca4e
+x9: dq 0xacbd382dcf5b8de2,0xd229e1f5b281303f
+x10: dq 0x71aeaff20b095fd9,0xab63e2e11fa38ed9
+x11: dq 0x89b0c0765892729a,0x77d410d5c42c882d
+x12: dq 0xc45ea11a955d8dd5,0x24b3c1d2a024048b
+x13: dq 0x2e8ec680de14b47c,0xdd7b8919edd42786
+x14: dq 0x135ce6888fa02cbf,0x11e53e2b2ac655ef
+x15: dq 0x011ff554472a7a10,0x6de8f4c914c334d5
+n7: dq 0x21f86d66c8ca00ce
+n8: dq 0x75b6ba21077c48ad
+n9: dq 0xed56bb2dcb3c7736
+n10: dq 0x8bda43d3fd1a7e06
+n11: dq 0xb64a9c9e5d318408
+n12: dq 0xdf9a54b303f1d3a3
+n13: dq 0x4a75479abd64e097
+n14: dq 0x249214109d5d1c88
%endif
SECTION .text
; (max_args % 4) must equal 3 for stack alignment
%define max_args 15
+%if ARCH_X86_64
+
+;-----------------------------------------------------------------------------
+; void x264_checkasm_stack_clobber( uint64_t clobber, ... )
+;-----------------------------------------------------------------------------
+cglobal checkasm_stack_clobber, 1,2
+ ; Clobber the stack with junk below the stack pointer
+ %define size (max_args+6)*8
+ SUB rsp, size
+ mov r1, size-8
+.loop:
+ mov [rsp+r1], r0
+ sub r1, 8
+ jge .loop
+ ADD rsp, size
+ RET
+
%if WIN64
+ %assign free_regs 7
+%else
+ %assign free_regs 9
+%endif
;-----------------------------------------------------------------------------
; intptr_t x264_checkasm_call( intptr_t (*func)(), int *ok, ... )
;-----------------------------------------------------------------------------
INIT_XMM
-cglobal checkasm_call, 4,15,16
- SUB rsp, max_args*8
+cglobal checkasm_call, 2,15,16,max_args*8+8
mov r6, r0
- mov [rsp+stack_offset+16], r1
- mov r0, r2
- mov r1, r3
- mov r2d, r4m ; FIXME truncates pointer
- mov r3d, r5m ; FIXME truncates pointer
-%assign i 4
-%rep max_args-4
- mov r4, [rsp+stack_offset+8+(i+2)*8]
- mov [rsp+i*8], r4
- %assign i i+1
-%endrep
-%assign i 6
-%rep 16-6
- mova m %+ i, [x %+ i]
- %assign i i+1
-%endrep
-%assign i 7
-%rep 15-7
+ mov [rsp+max_args*8], r1
+
+ ; All arguments have been pushed on the stack instead of registers in order to
+ ; test for incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
+ mov r0, r6mp
+ mov r1, r7mp
+ mov r2, r8mp
+ mov r3, r9mp
+%if UNIX64
+ mov r4, r10mp
+ mov r5, r11mp
+ %assign i 6
+ %rep max_args-6
+ mov r9, [rsp+stack_offset+(i+1)*8]
+ mov [rsp+(i-6)*8], r9
+ %assign i i+1
+ %endrep
+%else
+ %assign i 4
+ %rep max_args-4
+ mov r9, [rsp+stack_offset+(i+7)*8]
+ mov [rsp+i*8], r9
+ %assign i i+1
+ %endrep
+%endif
+
+%if WIN64
+ %assign i 6
+ %rep 16-6
+ mova m %+ i, [x %+ i]
+ %assign i i+1
+ %endrep
+%endif
+
+%assign i 14
+%rep 15-free_regs
mov r %+ i, [n %+ i]
- %assign i i+1
+ %assign i i-1
%endrep
call r6
-%assign i 7
-%rep 15-7
+%assign i 14
+%rep 15-free_regs
xor r %+ i, [n %+ i]
- or r7, r %+ i
- %assign i i+1
-%endrep
-%assign i 6
-%rep 16-6
- pxor m %+ i, [x %+ i]
- por m6, m %+ i
- %assign i i+1
+ or r14, r %+ i
+ %assign i i-1
%endrep
+
+%if WIN64
+ %assign i 6
+ %rep 16-6
+ pxor m %+ i, [x %+ i]
+ por m6, m %+ i
+ %assign i i+1
+ %endrep
packsswb m6, m6
movq r5, m6
- or r7, r5
+ or r14, r5
+%endif
+
jz .ok
- mov r4, rax
+ mov r9, rax
lea r0, [error_message]
call puts
- mov r1, [rsp+stack_offset+16]
+ mov r1, [rsp+max_args*8]
mov dword [r1], 0
- mov rax, r4
+ mov rax, r9
.ok:
- ADD rsp, max_args*8
RET
-%elif ARCH_X86_64 == 0
+%else
; just random numbers to reduce the chance of incidental match
%define n3 dword 0x6549315c
mov dword [r1], 0
mov eax, r3
.ok:
- RET
+ REP_RET
%endif ; ARCH_X86_64
; int x264_stack_pagealign( int (*func)(), int align )
;-----------------------------------------------------------------------------
cglobal stack_pagealign, 2,2
+ movsxdifnidn r1, r1d
push rbp
mov rbp, rsp
+%if WIN64
+ sub rsp, 32 ; shadow space
+%endif
and rsp, ~0xfff
sub rsp, r1
call r0
projects / x264 / blobdiff