/* Copyright (c) 2014, Intel Corporation All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * Neither the name of Intel Corporation nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /******************************************************************************/ // ALGORITHM DESCRIPTION // --------------------- // // X87 version: // Use 80-bit FPU precision fmul, fsqrt to compute square and sqrt. // // SSE version: // Swap x, y if |x|<|y| // For x=2^k*x, get y=y*2^(-k) // Get S ~ sqrt(x^2+y^2) (leading 1 + leading 25 mantissa bits) // // Get D = ( RN(x^2+y^2) - S^2 ) + ( x^2 - RN(x^2) ) + // + ( y^2 - ((RN(x^2+y^2)-RN(x^2)) ) // // Result is 2^k*(S + Se), where Se = S*e // S*e is approximated as (D/2S)*( 1 - (D/2S)^2*1.0/S ) // // Return 2^k*(S+Se) // // For |y/x|<2^(-64), return x // // For cases where maximum biased exponent is either greater than 7fdh or // below 32, take a special path to check for special cases (0, NaN, Inf), // possible overflow, and more accurate computation for denormal results // // Special cases: // hypot(x,y), hypot(y,x), and hypot(x,-y) are equivalent // hypot(x,+-0) is equivalent to fabs(x) // hypot(x,y) = y if (x==NaN or x==INF) and y==INF // hypot(x,y) = x if (x==NaN or x==INF) and y!=INF (even if y==NaN!) // hypot(x,y) = y if (x!=NaN and x!=INF) and (y==NaN or y==INF) // /******************************************************************************/ #include # -- Begin hypot ENTRY(hypot) # parameter 1: %xmm0 # parameter 2: %xmm1 ..B1.1: ..___tag_value_hypot.1: ..___tag_value_hypot.3: ..B1.2: subq $64, %rsp movapd static_const_table(%rip), %xmm3 movsd %xmm0, 48(%rsp) movsd %xmm1, 56(%rsp) andpd %xmm3, %xmm0 andpd %xmm3, %xmm1 pextrw $3, %xmm0, %eax pextrw $3, %xmm1, %edx cmpl $24528, %eax ja .L_2TAG_PACKET_0.0.1 cmpl $24528, %edx ja .L_2TAG_PACKET_0.0.1 .L_2TAG_PACKET_1.0.1: fldl 48(%rsp) fldl 56(%rsp) fxch %st(1) fmul %st(0), %st fxch %st(1) nop fmul %st(0), %st faddp %st, %st(1) fsqrt jmp .L_2TAG_PACKET_2.0.1 .L_2TAG_PACKET_0.0.1: cmpl $32752, %eax movl %eax, %ecx jae .L_2TAG_PACKET_3.0.1 subl %edx, %ecx cmpl $32752, %edx jae .L_2TAG_PACKET_3.0.1 addl $928, %ecx addl %edx, %eax cmpl $1856, %ecx ja .L_2TAG_PACKET_4.0.1 cmpl $49056, %eax jb .L_2TAG_PACKET_1.0.1 fldl 48(%rsp) fldl 56(%rsp) fxch %st(1) fmul %st(0), %st fxch %st(1) nop fmul %st(0), %st faddp %st, %st(1) fsqrt .L_2TAG_PACKET_5.0.1: fstl (%rsp) fstpt 16(%rsp) xorl %eax, %eax movw 24(%rsp), %ax cmpl $17407, %eax jae .L_2TAG_PACKET_6.0.1 fldl (%rsp) jmp .L_2TAG_PACKET_7.0.1 .L_2TAG_PACKET_4.0.1: movsd %xmm0, 32(%rsp) movsd %xmm1, 40(%rsp) fldl 32(%rsp) faddl 40(%rsp) jmp .L_2TAG_PACKET_5.0.1 .L_2TAG_PACKET_6.0.1: fldl (%rsp) jmp .L_2TAG_PACKET_7.0.1 .L_2TAG_PACKET_3.0.1: shufpd $0, %xmm1, %xmm0 movdqa %xmm0, %xmm2 movdqa 16+static_const_table(%rip), %xmm3 movsd %xmm0, 32(%rsp) movsd %xmm1, 40(%rsp) cmppd $3, %xmm0, %xmm2 cmppd $0, %xmm0, %xmm3 movmskpd %xmm2, %edx movmskpd %xmm3, %rax testl %edx, %edx je .L_2TAG_PACKET_8.0.1 fldl 32(%rsp) fmull 40(%rsp) testq $1, %rax jne .L_2TAG_PACKET_9.0.1 testq $2, %rax jne .L_2TAG_PACKET_10.0.1 jmp .L_2TAG_PACKET_2.0.1 .L_2TAG_PACKET_8.0.1: fldl 32(%rsp) faddl 40(%rsp) jmp .L_2TAG_PACKET_2.0.1 .L_2TAG_PACKET_9.0.1: fstpl 40(%rsp) fldl 32(%rsp) jmp .L_2TAG_PACKET_7.0.1 .L_2TAG_PACKET_10.0.1: fstpl 32(%rsp) fldl 40(%rsp) jmp .L_2TAG_PACKET_7.0.1 .L_2TAG_PACKET_2.0.1: .L_2TAG_PACKET_7.0.1: fstpl 16(%rsp) movq 16(%rsp), %xmm0 addq $64, %rsp ret ..B1.3: ..___tag_value_hypot.4: END(hypot) # -- End hypot .section .rodata, "a" .align 16 .align 16 static_const_table: .long 4294967295 .long 2147483647 .long 4294967295 .long 2147483647 .long 0 .long 2146435072 .long 0 .long 2146435072 .type static_const_table,@object .size static_const_table,32 .data .section .note.GNU-stack, "",@progbits // -- Begin DWARF2 SEGMENT .eh_frame .section .eh_frame,"a",@progbits .eh_frame_seg: .align 1 .4byte 0x00000014 .8byte 0x00527a0100000000 .8byte 0x08070c1b01107801 .4byte 0x00000190 .4byte 0x00000014 .4byte 0x0000001c .4byte ..___tag_value_hypot.1-. .4byte ..___tag_value_hypot.4-..___tag_value_hypot.1 .2byte 0x0400 .4byte ..___tag_value_hypot.3-..___tag_value_hypot.1 .2byte 0x100e # End