diff options
Diffstat (limited to 'sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S')
| -rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S | 212 |
1 files changed, 212 insertions, 0 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S new file mode 100644 index 0000000000..a34e267433 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S @@ -0,0 +1,212 @@ +/* Function exp vectorized with AVX2. + Copyright (C) 2014-2015 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 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 + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + <http://www.gnu.org/licenses/>. */ + +#include <sysdep.h> +#include "svml_d_exp_data.h" + + .text +ENTRY (_ZGVdN4v_exp_avx2) +/* + ALGORITHM DESCRIPTION: + + Argument representation: + N = rint(X*2^k/ln2) = 2^k*M+j + X = N*ln2/2^k + r = M*ln2 + ln2*(j/2^k) + r + then -ln2/2^(k+1) < r < ln2/2^(k+1) + Alternatively: + N = trunc(X*2^k/ln2) + then 0 < r < ln2/2^k + + Result calculation: + exp(X) = exp(M*ln2 + ln2*(j/2^k) + r) + = 2^M * 2^(j/2^k) * exp(r) + 2^M is calculated by bit manipulation + 2^(j/2^k) is stored in table + exp(r) is approximated by polynomial + + The table lookup is skipped if k = 0. */ + + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $448, %rsp + movq __svml_dexp_data@GOTPCREL(%rip), %rax + vmovdqa %ymm0, %ymm2 + vmovupd __dbInvLn2(%rax), %ymm3 + vmovupd __dbShifter(%rax), %ymm1 + vmovupd __lIndexMask(%rax), %ymm4 + +/* dM = X*dbInvLn2+dbShifter, dbInvLn2 = 2^k/Ln2 */ + vfmadd213pd %ymm1, %ymm2, %ymm3 + +/* iAbsX = (int)(lX>>32), lX = *(longlong*)&X */ + vextracti128 $1, %ymm2, %xmm5 + vshufps $221, %xmm5, %xmm2, %xmm6 + +/* iAbsX = iAbsX&iAbsMask */ + vandps __iAbsMask(%rax), %xmm6, %xmm7 + +/* dN = dM-dbShifter, dN = rint(X*2^k/Ln2) */ + vsubpd %ymm1, %ymm3, %ymm6 + +/* iRangeMask = (iAbsX>iDomainRange) */ + vpcmpgtd __iDomainRange(%rax), %xmm7, %xmm0 + vmovupd __dbLn2hi(%rax), %ymm1 + vmovupd __dPC0(%rax), %ymm7 + +/* Mask = iRangeMask?1:0, set mask for overflow/underflow */ + vmovmskps %xmm0, %ecx + vmovupd __dPC2(%rax), %ymm0 + +/* dR = X - dN*dbLn2hi, dbLn2hi is 52-8-k hi bits of ln2/2^k */ + vmovdqa %ymm2, %ymm5 + vfnmadd231pd %ymm6, %ymm1, %ymm5 + +/* dR = dR - dN*dbLn2lo, dbLn2lo is 40..94 bits of lo part of ln2/2^k */ + vfnmadd132pd __dbLn2lo(%rax), %ymm5, %ymm6 + +/* exp(r) = b0+r*(b0+r*(b1+r*b2)) */ + vfmadd213pd __dPC1(%rax), %ymm6, %ymm0 + vfmadd213pd %ymm7, %ymm6, %ymm0 + vfmadd213pd %ymm7, %ymm6, %ymm0 + +/* lIndex = (*(longlong*)&dM)&lIndexMask, lIndex is the lower K bits of lM */ + vandps %ymm4, %ymm3, %ymm1 + +/* table lookup for dT[j] = 2^(j/2^k) */ + vxorpd %ymm6, %ymm6, %ymm6 + vpcmpeqd %ymm5, %ymm5, %ymm5 + vgatherqpd %ymm5, (%rax,%ymm1,8), %ymm6 + +/* lM = (*(longlong*)&dM)&(~lIndexMask) */ + vpandn %ymm3, %ymm4, %ymm3 + +/* 2^(j/2^k) * exp(r) */ + vmulpd %ymm0, %ymm6, %ymm0 + +/* lM = lM<<(52-K), 2^M */ + vpsllq $42, %ymm3, %ymm4 + +/* multiply by 2^M through integer add */ + vpaddq %ymm4, %ymm0, %ymm0 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_1_3: + cfi_restore_state + vmovupd %ymm2, 320(%rsp) + vmovupd %ymm0, 384(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + xorl %eax, %eax + vmovups %ymm8, 224(%rsp) + vmovups %ymm9, 192(%rsp) + vmovups %ymm10, 160(%rsp) + vmovups %ymm11, 128(%rsp) + vmovups %ymm12, 96(%rsp) + vmovups %ymm13, 64(%rsp) + vmovups %ymm14, 32(%rsp) + vmovups %ymm15, (%rsp) + movq %rsi, 264(%rsp) + movq %rdi, 256(%rsp) + movq %r12, 296(%rsp) + cfi_offset_rel_rsp (12, 296) + movb %dl, %r12b + movq %r13, 288(%rsp) + cfi_offset_rel_rsp (13, 288) + movl %ecx, %r13d + movq %r14, 280(%rsp) + cfi_offset_rel_rsp (14, 280) + movl %eax, %r14d + movq %r15, 272(%rsp) + cfi_offset_rel_rsp (15, 272) + cfi_remember_state + +.LBL_1_6: + btl %r14d, %r13d + jc .LBL_1_12 + +.LBL_1_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_1_10 + +.LBL_1_8: + incb %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_1_6 + + vmovups 224(%rsp), %ymm8 + vmovups 192(%rsp), %ymm9 + vmovups 160(%rsp), %ymm10 + vmovups 128(%rsp), %ymm11 + vmovups 96(%rsp), %ymm12 + vmovups 64(%rsp), %ymm13 + vmovups 32(%rsp), %ymm14 + vmovups (%rsp), %ymm15 + vmovupd 384(%rsp), %ymm0 + movq 264(%rsp), %rsi + movq 256(%rsp), %rdi + movq 296(%rsp), %r12 + cfi_restore (%r12) + movq 288(%rsp), %r13 + cfi_restore (%r13) + movq 280(%rsp), %r14 + cfi_restore (%r14) + movq 272(%rsp), %r15 + cfi_restore (%r15) + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 328(%rsp,%r15), %xmm0 + vzeroupper + + call exp@PLT + + vmovsd %xmm0, 392(%rsp,%r15) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 320(%rsp,%r15), %xmm0 + vzeroupper + + call exp@PLT + + vmovsd %xmm0, 384(%rsp,%r15) + jmp .LBL_1_7 + +END (_ZGVdN4v_exp_avx2) |