diff options
author | André Fabian Silva Delgado <emulatorman@parabola.nu> | 2015-08-05 17:04:01 -0300 |
---|---|---|
committer | André Fabian Silva Delgado <emulatorman@parabola.nu> | 2015-08-05 17:04:01 -0300 |
commit | 57f0f512b273f60d52568b8c6b77e17f5636edc0 (patch) | |
tree | 5e910f0e82173f4ef4f51111366a3f1299037a7b /arch/x86/math-emu/wm_shrx.S |
Initial import
Diffstat (limited to 'arch/x86/math-emu/wm_shrx.S')
-rw-r--r-- | arch/x86/math-emu/wm_shrx.S | 204 |
1 files changed, 204 insertions, 0 deletions
diff --git a/arch/x86/math-emu/wm_shrx.S b/arch/x86/math-emu/wm_shrx.S new file mode 100644 index 000000000..518428317 --- /dev/null +++ b/arch/x86/math-emu/wm_shrx.S @@ -0,0 +1,204 @@ + .file "wm_shrx.S" +/*---------------------------------------------------------------------------+ + | wm_shrx.S | + | | + | 64 bit right shift functions | + | | + | Copyright (C) 1992,1995 | + | W. Metzenthen, 22 Parker St, Ormond, Vic 3163, | + | Australia. E-mail billm@jacobi.maths.monash.edu.au | + | | + | Call from C as: | + | unsigned FPU_shrx(void *arg1, unsigned arg2) | + | and | + | unsigned FPU_shrxs(void *arg1, unsigned arg2) | + | | + +---------------------------------------------------------------------------*/ + +#include "fpu_emu.h" + +.text +/*---------------------------------------------------------------------------+ + | unsigned FPU_shrx(void *arg1, unsigned arg2) | + | | + | Extended shift right function. | + | Fastest for small shifts. | + | Shifts the 64 bit quantity pointed to by the first arg (arg1) | + | right by the number of bits specified by the second arg (arg2). | + | Forms a 96 bit quantity from the 64 bit arg and eax: | + | [ 64 bit arg ][ eax ] | + | shift right ---------> | + | The eax register is initialized to 0 before the shifting. | + | Results returned in the 64 bit arg and eax. | + +---------------------------------------------------------------------------*/ + +ENTRY(FPU_shrx) + push %ebp + movl %esp,%ebp + pushl %esi + movl PARAM2,%ecx + movl PARAM1,%esi + cmpl $32,%ecx /* shrd only works for 0..31 bits */ + jnc L_more_than_31 + +/* less than 32 bits */ + pushl %ebx + movl (%esi),%ebx /* lsl */ + movl 4(%esi),%edx /* msl */ + xorl %eax,%eax /* extension */ + shrd %cl,%ebx,%eax + shrd %cl,%edx,%ebx + shr %cl,%edx + movl %ebx,(%esi) + movl %edx,4(%esi) + popl %ebx + popl %esi + leave + ret + +L_more_than_31: + cmpl $64,%ecx + jnc L_more_than_63 + + subb $32,%cl + movl (%esi),%eax /* lsl */ + movl 4(%esi),%edx /* msl */ + shrd %cl,%edx,%eax + shr %cl,%edx + movl %edx,(%esi) + movl $0,4(%esi) + popl %esi + leave + ret + +L_more_than_63: + cmpl $96,%ecx + jnc L_more_than_95 + + subb $64,%cl + movl 4(%esi),%eax /* msl */ + shr %cl,%eax + xorl %edx,%edx + movl %edx,(%esi) + movl %edx,4(%esi) + popl %esi + leave + ret + +L_more_than_95: + xorl %eax,%eax + movl %eax,(%esi) + movl %eax,4(%esi) + popl %esi + leave + ret + + +/*---------------------------------------------------------------------------+ + | unsigned FPU_shrxs(void *arg1, unsigned arg2) | + | | + | Extended shift right function (optimized for small floating point | + | integers). | + | Shifts the 64 bit quantity pointed to by the first arg (arg1) | + | right by the number of bits specified by the second arg (arg2). | + | Forms a 96 bit quantity from the 64 bit arg and eax: | + | [ 64 bit arg ][ eax ] | + | shift right ---------> | + | The eax register is initialized to 0 before the shifting. | + | The lower 8 bits of eax are lost and replaced by a flag which is | + | set (to 0x01) if any bit, apart from the first one, is set in the | + | part which has been shifted out of the arg. | + | Results returned in the 64 bit arg and eax. | + +---------------------------------------------------------------------------*/ +ENTRY(FPU_shrxs) + push %ebp + movl %esp,%ebp + pushl %esi + pushl %ebx + movl PARAM2,%ecx + movl PARAM1,%esi + cmpl $64,%ecx /* shrd only works for 0..31 bits */ + jnc Ls_more_than_63 + + cmpl $32,%ecx /* shrd only works for 0..31 bits */ + jc Ls_less_than_32 + +/* We got here without jumps by assuming that the most common requirement + is for small integers */ +/* Shift by [32..63] bits */ + subb $32,%cl + movl (%esi),%eax /* lsl */ + movl 4(%esi),%edx /* msl */ + xorl %ebx,%ebx + shrd %cl,%eax,%ebx + shrd %cl,%edx,%eax + shr %cl,%edx + orl %ebx,%ebx /* test these 32 bits */ + setne %bl + test $0x7fffffff,%eax /* and 31 bits here */ + setne %bh + orw %bx,%bx /* Any of the 63 bit set ? */ + setne %al + movl %edx,(%esi) + movl $0,4(%esi) + popl %ebx + popl %esi + leave + ret + +/* Shift by [0..31] bits */ +Ls_less_than_32: + movl (%esi),%ebx /* lsl */ + movl 4(%esi),%edx /* msl */ + xorl %eax,%eax /* extension */ + shrd %cl,%ebx,%eax + shrd %cl,%edx,%ebx + shr %cl,%edx + test $0x7fffffff,%eax /* only need to look at eax here */ + setne %al + movl %ebx,(%esi) + movl %edx,4(%esi) + popl %ebx + popl %esi + leave + ret + +/* Shift by [64..95] bits */ +Ls_more_than_63: + cmpl $96,%ecx + jnc Ls_more_than_95 + + subb $64,%cl + movl (%esi),%ebx /* lsl */ + movl 4(%esi),%eax /* msl */ + xorl %edx,%edx /* extension */ + shrd %cl,%ebx,%edx + shrd %cl,%eax,%ebx + shr %cl,%eax + orl %ebx,%edx + setne %bl + test $0x7fffffff,%eax /* only need to look at eax here */ + setne %bh + orw %bx,%bx + setne %al + xorl %edx,%edx + movl %edx,(%esi) /* set to zero */ + movl %edx,4(%esi) /* set to zero */ + popl %ebx + popl %esi + leave + ret + +Ls_more_than_95: +/* Shift by [96..inf) bits */ + xorl %eax,%eax + movl (%esi),%ebx + orl 4(%esi),%ebx + setne %al + xorl %ebx,%ebx + movl %ebx,(%esi) + movl %ebx,4(%esi) + popl %ebx + popl %esi + leave + ret |