/* Compute cubic root of long double value. Copyright (C) 1997 Free Software Foundation, Inc. This file is part of the GNU C Library. Contributed by Dirk Alboth and Ulrich Drepper , 1997. The GNU C Library is free software; you can redistribute it and/or modify it under the terms of the GNU Library General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. The GNU C Library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public License for more details. You should have received a copy of the GNU Library General Public License along with the GNU C Library; see the file COPYING.LIB. If not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ #include #ifdef __ELF__ .section .rodata #else .text #endif .align ALIGNARG(4) ASM_TYPE_DIRECTIVE(f1,@object) f1: .double 0.338058687610520237 ASM_SIZE_DIRECTIVE(f1) ASM_TYPE_DIRECTIVE(f2,@object) f2: .double 1.67595307700780102 ASM_SIZE_DIRECTIVE(f2) ASM_TYPE_DIRECTIVE(f3,@object) f3: .double -2.82414939754975962 ASM_SIZE_DIRECTIVE(f3) ASM_TYPE_DIRECTIVE(f4,@object) f4: .double 4.09559907378707839 ASM_SIZE_DIRECTIVE(f4) ASM_TYPE_DIRECTIVE(f5,@object) f5: .double -4.11151425200350531 ASM_SIZE_DIRECTIVE(f5) ASM_TYPE_DIRECTIVE(f6,@object) f6: .double 2.65298938441952296 ASM_SIZE_DIRECTIVE(f6) ASM_TYPE_DIRECTIVE(f7,@object) f7: .double -0.988553671195413709 ASM_SIZE_DIRECTIVE(f7) ASM_TYPE_DIRECTIVE(f8,@object) f8: .double 0.161617097923756032 ASM_SIZE_DIRECTIVE(f8) #define CBRT2 1.2599210498948731648 #define SQR_CBRT2 1.5874010519681994748 ASM_TYPE_DIRECTIVE(factor,@object) factor: .double 1.0 / SQR_CBRT2 .double 1.0 / CBRT2 .double 1.0 .double CBRT2 .double SQR_CBRT2 ASM_SIZE_DIRECTIVE(factor) ASM_TYPE_DIRECTIVE(two64,@object) two64: .byte 0, 0, 0, 0, 0, 0, 0xf0, 0x43 ASM_SIZE_DIRECTIVE(two64) #ifdef PIC #define MO(op) op##@GOTOFF(%ebx) #define MOX(op,x,f) op##@GOTOFF(%ebx,x,f) #else #define MO(op) op #define MOX(op,x,f) op(,x,f) #endif .text ENTRY(__cbrtl) movl 4(%esp), %ecx movl 12(%esp), %eax orl 8(%esp), %ecx movl %eax, %edx andl $0x7fff, %eax orl %eax, %ecx jz 1f xorl %ecx, %ecx cmpl $0x7fff, %eax je 1f #ifdef PIC pushl %ebx call 3f 3: popl %ebx addl $_GLOBAL_OFFSET_TABLE_+[.-3b], %ebx #endif cmpl $0, %eax je 2f #ifdef PIC fldt 8(%esp) #else fldt 4(%esp) #endif fmull MO(two64) movl $-64, %ecx fstpt 4(%esp) movl 12(%esp), %eax movl %eax, %edx andl $0x7fff, %eax 2: andl $0x8000, %edx subl $16382, %eax orl $0x3ffe, %edx addl %eax, %ecx #ifdef PIC movl %edx, 16(%esp) fldt 8(%esp) /* xm */ #else movl %edx, 12(%esp) fldt 4(%esp) /* xm */ #endif fabs /* The following code has two track: a) compute the normalized cbrt value b) compute xe/3 and xe%3 The right track computes the value for b) and this is done in an optimized way by avoiding division. */ fld %st(0) /* xm : xm */ fmull MO(f7) /* f7*xm : xm */ movl $1431655766, %eax faddl MO(f6) /* f6+f7*xm : xm */ imull %ecx fmul %st(1) /* (f6+f7*xm)*xm : xm */ movl %ecx, %eax faddl MO(f5) /* f5+(f6+f7*xm)*xm : xm */ sarl $31, %eax fmul %st(1) /* (f5+(f6+f7*xm)*xm)*xm : xm */ subl %eax, %edx faddl MO(f4) /* f4+(f5+(f6+f7*xm)*xm)*xm : xm */ fmul %st(1) /* (f4+(f5+(f6+f7*xm)*xm)*xm)*xm : xm */ faddl MO(f3) /* f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm : xm */ fmul %st(1) /* (f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm : xm */ faddl MO(f2) /* f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm : xm */ fmul %st(1) /* (f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm)*xm : xm */ faddl MO(f1) /* u:=f1+(f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm)*xm : xm */ fld %st /* u : u : xm */ fmul %st(1) /* u*u : u : xm */ fld %st(2) /* xm : u*u : u : xm */ fadd %st /* 2*xm : u*u : u : xm */ fxch %st(1) /* u*u : 2*xm : u : xm */ fmul %st(2) /* t2:=u*u*u : 2*xm : u : xm */ movl %edx, %eax fadd %st, %st(1) /* t2 : t2+2*xm : u : xm */ leal (%edx,%edx,2),%edx fadd %st(0) /* 2*t2 : t2+2*xm : u : xm */ subl %edx, %ecx faddp %st, %st(3) /* t2+2*xm : u : 2*t2+xm */ fmulp /* u*(t2+2*xm) : 2*t2+xm */ fdivp %st, %st(1) /* u*(t2+2*xm)/(2*t2+xm) */ fmull MOX(16+factor,%ecx,8) /* u*(t2+2*xm)/(2*t2+xm)*FACT */ pushl %eax fildl (%esp) /* xe/3 : u*(t2+2*xm)/(2*t2+xm)*FACT */ fxch /* u*(t2+2*xm)/(2*t2+xm)*FACT : xe/3 */ popl %eax fscale /* u*(t2+2*xm)/(2*t2+xm)*FACT*2^xe/3 */ fstp %st(1) #ifdef PIC popl %ebx #endif testl $0x8000, 12(%esp) jz 4f fchs 4: ret /* Return the argument. */ 1: fldt 4(%esp) ret END(__cbrtl) weak_alias (__cbrtl, cbrtl)