diff options
| author | Rokas Kupstys <rokups@zoho.com> | 2017-02-20 17:10:50 +0200 |
|---|---|---|
| committer | Rokas Kupstys <rokups@zoho.com> | 2017-02-20 17:24:19 +0200 |
| commit | 373e667dbc6aff5dd74f5ada139d867287e9dc1f (patch) | |
| tree | 41da1a7c7298e8ae5f146b528446b3d279480aca /lib/arch/x86/amd64.S | |
| parent | a3b8bf300df20f2922275ff62293e3f882cad090 (diff) | |
| download | Nim-373e667dbc6aff5dd74f5ada139d867287e9dc1f.tar.gz | |
Coroutine rework.
* ucontext backend (default on unix)
* setjmp backend
* fibers backend (default and required on windows)
* Fixed coroutine loop timing issues
* Fixed saving of xmm registers on x64 windows
* Fixed alignment issues
* Updated coroutine sample with cooperative fibonacci calculation.
* Disable glibc security features only when platform jump functions are used
* Removed dependency on fasm.
* Using fiber api on windows.
* Other platforms and compilers will use built in assembler and .S files or API provided by platform libc.
* Replaced stack switching procs with `coroExecWithStack()` which never returns. This makes compiler always generate proper code.
Diffstat (limited to 'lib/arch/x86/amd64.S')
| -rw-r--r-- | lib/arch/x86/amd64.S | 96 |
1 files changed, 96 insertions, 0 deletions
diff --git a/lib/arch/x86/amd64.S b/lib/arch/x86/amd64.S new file mode 100644 index 000000000..47a26f627 --- /dev/null +++ b/lib/arch/x86/amd64.S @@ -0,0 +1,96 @@ +# +# +# Nim's Runtime Library +# (c) Copyright 2015 Rokas Kupstys +# +# See the file "copying.txt", included in this +# distribution, for details about the copyright. +# +# Partially based on code from musl libc Copyright © 2005-2014 Rich Felker, et al. + +.globl narch_coroExecWithStack +.globl narch_setjmp +.globl narch_longjmp +.text + + +# SysV ABI - first argument is rdi. +# MS ABI - first argument is rcx. +#if defined(__MINGW32__) || defined(__MINGW64__) + #define REG_ARG1 rcx + #define REG_ARG2 rdx +#else + #define REG_ARG1 rdi + #define REG_ARG2 rsi +#endif + + +narch_coroExecWithStack: + mov %REG_ARG2, %rsp # swap stack with one passed to func + sub $0x30, %rsp # shadow space (for ms ABI) 0x20 + 0x10 for possible misalignment + and $-0x10, %rsp # 16-byte stack alignment + call *%REG_ARG1 + + +narch_setjmp: + add $0x10, %REG_ARG1 # 16-byte alignment + and $-0x10, %REG_ARG1 + mov %rbx, 0x00(%REG_ARG1) # jmp_buf, move registers onto it + mov %rbp, 0x08(%REG_ARG1) + mov %r12, 0x10(%REG_ARG1) + mov %r13, 0x18(%REG_ARG1) + mov %r14, 0x20(%REG_ARG1) + mov %r15, 0x28(%REG_ARG1) + lea 0x08(%rsp), %rdx # this is our rsp WITHOUT current ret addr + mov %rdx, 0x30(%REG_ARG1) + mov (%rsp), %rdx # save return addr ptr for new rip + mov %rdx, 0x38(%REG_ARG1) + mov %rsi, 0x40(%REG_ARG1) + mov %rdi, 0x48(%REG_ARG1) +#if defined(__MINGW32__) || defined(__MINGW64__) + movaps %xmm6, 0x50(%REG_ARG1) + movaps %xmm7, 0x60(%REG_ARG1) + movaps %xmm8, 0x70(%REG_ARG1) + movaps %xmm9, 0x80(%REG_ARG1) + movaps %xmm10, 0x90(%REG_ARG1) + movaps %xmm11, 0xA0(%REG_ARG1) + movaps %xmm12, 0xB0(%REG_ARG1) + movaps %xmm13, 0xC0(%REG_ARG1) + movaps %xmm14, 0xD0(%REG_ARG1) + movaps %xmm15, 0xE0(%REG_ARG1) +#endif + xor %rax, %rax # always return 0 + ret + + +narch_longjmp: + add $0x10, %REG_ARG1 # 16-byte alignment + and $-0x10, %REG_ARG1 # + mov %REG_ARG2, %rax # val will be longjmp return + test %rax, %rax + jnz narch_longjmp_1 + inc %rax # if val==0, val=1 per longjmp semantics +narch_longjmp_1: + mov 0x00(%REG_ARG1), %rbx # jmp_buf, restore regs from it + mov 0x08(%REG_ARG1), %rbp + mov 0x10(%REG_ARG1), %r12 + mov 0x18(%REG_ARG1), %r13 + mov 0x20(%REG_ARG1), %r14 + mov 0x28(%REG_ARG1), %r15 + mov 0x30(%REG_ARG1), %rsp # this ends up being the stack pointer + mov 0x38(%REG_ARG1), %rdx # this is the instruction pointer + mov 0x40(%REG_ARG1), %rsi + mov 0x48(%REG_ARG1), %rdi +#if defined(__MINGW32__) || defined(__MINGW64__) + movaps 0x50(%REG_ARG1), %xmm6 + movaps 0x60(%REG_ARG1), %xmm7 + movaps 0x70(%REG_ARG1), %xmm8 + movaps 0x80(%REG_ARG1), %xmm9 + movaps 0x90(%REG_ARG1), %xmm10 + movaps 0xA0(%REG_ARG1), %xmm11 + movaps 0xB0(%REG_ARG1), %xmm12 + movaps 0xC0(%REG_ARG1), %xmm13 + movaps 0xD0(%REG_ARG1), %xmm14 + movaps 0xE0(%REG_ARG1), %xmm15 +#endif + jmp *%rdx # goto saved address without altering rsp |