diff options
Diffstat (limited to 'sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S')
-rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S | 741 |
1 files changed, 741 insertions, 0 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S new file mode 100644 index 0000000000..8dd89c8ebb --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S @@ -0,0 +1,741 @@ +/* Function pow vectorized with AVX-512. KNL and SKX versions. + 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_pow_data.h" +#include "svml_d_wrapper_impl.h" + +/* ALGORITHM DESCRIPTION: + + 1) Calculating log2|x| + Here we use the following formula. + Let |x|=2^k1*X1, where k1 is integer, 1<=X1<2. + Let C ~= 1/ln(2), + Rcp1 ~= 1/X1, X2=Rcp1*X1, + Rcp2 ~= 1/X2, X3=Rcp2*X2, + Rcp3 ~= 1/X3, Rcp3C ~= C/X3. + Then + log2|x| = k1 + log2(1/Rcp1) + log2(1/Rcp2) + log2(C/Rcp3C) + + log2(X1*Rcp1*Rcp2*Rcp3C/C), + where X1*Rcp1*Rcp2*Rcp3C = C*(1+q), q is very small. + + The values of Rcp1, log2(1/Rcp1), Rcp2, log2(1/Rcp2), + Rcp3C, log2(C/Rcp3C) are taken from tables. + Values of Rcp1, Rcp2, Rcp3C are such that RcpC=Rcp1*Rcp2*Rcp3C + is exactly represented in target precision. + + log2(X1*Rcp1*Rcp2*Rcp3C/C) = log2(1+q) = ln(1+q)/ln2 = + = 1/(ln2)*q - 1/(2ln2)*q^2 + 1/(3ln2)*q^3 - ... = + = 1/(C*ln2)*cq - 1/(2*C^2*ln2)*cq^2 + 1/(3*C^3*ln2)*cq^3 - ... = + = (1 + a1)*cq + a2*cq^2 + a3*cq^3 + ..., + where cq = X1*Rcp1*Rcp2*Rcp3C-C, + a1=1/(C*ln(2))-1 is small, + a2=1/(2*C^2*ln2), + a3=1/(3*C^3*ln2), + ... + We get 3 parts of log2 result: HH+HL+HLL ~= log2|x|. + + 2) Calculation of y*(HH+HL+HLL). + Split y into YHi+YLo. + Get high PH and medium PL parts of y*log2|x|. + Get low PLL part of y*log2|x|. + Now we have PH+PL+PLL ~= y*log2|x|. + + 3) Calculation of 2^(PH+PL+PLL). + Mathematical idea of computing 2^(PH+PL+PLL) is the following. + Let's represent PH+PL+PLL in the form N + j/2^expK + Z, + where expK=7 in this implementation, N and j are integers, + 0<=j<=2^expK-1, |Z|<2^(-expK-1). + Hence 2^(PH+PL+PLL) ~= 2^N * 2^(j/2^expK) * 2^Z, + where 2^(j/2^expK) is stored in a table, and + 2^Z ~= 1 + B1*Z + B2*Z^2 ... + B5*Z^5. + + We compute 2^(PH+PL+PLL) as follows. + Break PH into PHH + PHL, where PHH = N + j/2^expK. + Z = PHL + PL + PLL + Exp2Poly = B1*Z + B2*Z^2 ... + B5*Z^5 + Get 2^(j/2^expK) from table in the form THI+TLO. + Now we have 2^(PH+PL+PLL) ~= 2^N * (THI + TLO) * (1 + Exp2Poly). + + Get significand of 2^(PH+PL+PLL) in the form ResHi+ResLo: + ResHi := THI + ResLo := THI * Exp2Poly + TLO + + Get exponent ERes of the result: + Res := ResHi + ResLo: + Result := ex(Res) + N. */ + + .text +ENTRY (_ZGVeN8vv_pow_knl) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_pow +#else + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1344, %rsp + vpsrlq $32, %zmm0, %zmm13 + vmovaps %zmm1, %zmm12 + movq __svml_dpow_data@GOTPCREL(%rip), %rax + movl $255, %edx + vpmovqd %zmm13, %ymm10 + vpsrlq $32, %zmm12, %zmm14 + kmovw %edx, %k1 + movl $-1, %ecx + vpmovqd %zmm14, %ymm15 + +/* x1 = x; Hi(x1) = (Hi(x1)&0x000fffff)|0x3ff00000 */ + vmovups _dbOne(%rax), %zmm6 + +/* i = (((Hi(x) & 0x000ffe00) + 0x00000200) >> 10); -> i = (b1..b11 + 1) / 2 */ + vmovaps %zmm10, %zmm5 + +/* k = Hi(x); k = k - 0x3fe7fe00; k = k >> 20 */ + vpsubd _i3fe7fe00(%rax), %zmm10, %zmm14{%k1} + vpandd _iIndexMask(%rax), %zmm10, %zmm5{%k1} + vpsrad $20, %zmm14, %zmm14{%k1} + vpxord %zmm9, %zmm9, %zmm9 + vpaddd _HIDELTA(%rax), %zmm10, %zmm3{%k1} + vpaddd _iIndexAdd(%rax), %zmm5, %zmm5{%k1} + vpxord %zmm7, %zmm7, %zmm7 + vpaddd _i2p20_2p19(%rax), %zmm14, %zmm14{%k1} + vpcmpd $1, _LORANGE(%rax), %zmm3, %k2{%k1} + vpsrld $10, %zmm5, %zmm5{%k1} + vpandd _ABSMASK(%rax), %zmm15, %zmm2{%k1} + vpbroadcastd %ecx, %zmm1{%k2}{z} + +/* Index for reciprocal table */ + vpslld $3, %zmm5, %zmm8{%k1} + kxnorw %k2, %k2, %k2 + vgatherdpd 11712(%rax,%ymm8), %zmm9{%k2} + vpmovzxdq %ymm14, %zmm10 + +/* Index for log2 table */ + vpslld $4, %zmm5, %zmm13{%k1} + kxnorw %k2, %k2, %k2 + vpsllq $32, %zmm10, %zmm3 + vpxord %zmm8, %zmm8, %zmm8 + vpcmpd $5, _INF(%rax), %zmm2, %k3{%k1} + vpbroadcastd %ecx, %zmm4{%k3}{z} + vpternlogq $248, _iMantissaMask(%rax), %zmm0, %zmm6 + kxnorw %k3, %k3, %k3 + vpternlogq $168, _iffffffff00000000(%rax), %zmm10, %zmm3 + +/* x1Hi=x1; Lo(x1Hi)&=0xf8000000; x1Lo = x1-x1Hi */ + vpandq _iHighMask(%rax), %zmm6, %zmm2 + vgatherdpd 19976(%rax,%ymm13), %zmm8{%k2} + vpord %zmm4, %zmm1, %zmm11{%k1} + vsubpd _db2p20_2p19(%rax), %zmm3, %zmm1 + vsubpd %zmm2, %zmm6, %zmm5 + +/* r1 = x1*rcp1 */ + vmulpd %zmm9, %zmm6, %zmm6 + vgatherdpd 19968(%rax,%ymm13), %zmm7{%k3} + +/* cq = c+r1 */ + vaddpd _LHN(%rax), %zmm6, %zmm4 + +/* E = -r1+__fence(x1Hi*rcp1) */ + vfmsub213pd %zmm6, %zmm9, %zmm2 + +/* T = k + L1hi */ + vaddpd %zmm7, %zmm1, %zmm7 + +/* E=E+x1Lo*rcp1 */ + vfmadd213pd %zmm2, %zmm9, %zmm5 + +/* T_Rh = T + cq */ + vaddpd %zmm4, %zmm7, %zmm3 + +/* Rl = T-T_Rh; -> -Rh */ + vsubpd %zmm3, %zmm7, %zmm9 + +/* Rl=Rl+cq */ + vaddpd %zmm9, %zmm4, %zmm6 + +/* T_Rh_Eh = T_Rh + E */ + vaddpd %zmm5, %zmm3, %zmm9 + +/* HLL = T_Rh - T_Rh_Eh; -> -Eh */ + vsubpd %zmm9, %zmm3, %zmm2 + +/* cq = cq + E; */ + vaddpd %zmm5, %zmm4, %zmm4 + +/* HLL+=E; -> El */ + vaddpd %zmm2, %zmm5, %zmm1 + vmovups _clv_2(%rax), %zmm5 + +/* HLL = HLL + (((((((a7)*cq+a6)*cq+a5)*cq+a4)*cq+a3)*cq+a2)*cq+a1)*cq */ + vfmadd213pd _clv_3(%rax), %zmm4, %zmm5 + +/* HLL+=Rl */ + vaddpd %zmm6, %zmm1, %zmm7 + +/* 2^(y*(HH+HL+HLL)) starts here: + yH = y; Lo(yH)&=0xf8000000 + */ + vpandq _iHighMask(%rax), %zmm12, %zmm6 + +/* yL = y-yH */ + vsubpd %zmm6, %zmm12, %zmm2 + vfmadd213pd _clv_4(%rax), %zmm4, %zmm5 + +/* HLL+=L1lo */ + vaddpd %zmm8, %zmm7, %zmm8 + vfmadd213pd _clv_5(%rax), %zmm4, %zmm5 + vfmadd213pd _clv_6(%rax), %zmm4, %zmm5 + vfmadd213pd _clv_7(%rax), %zmm4, %zmm5 + vfmadd213pd %zmm8, %zmm4, %zmm5 + +/* T_Rh_Eh_HLLhi = T_Rh_Eh + HLL */ + vaddpd %zmm5, %zmm9, %zmm13 + +/* HLLhi = T_Rh_Eh_HLLhi - T_Rh_Eh */ + vsubpd %zmm9, %zmm13, %zmm10 + +/* HLL = HLL - HLLhi */ + vsubpd %zmm10, %zmm5, %zmm3 + +/* HH = T_Rh_Eh_HLLhi; Lo(HH)&=0xf8000000 */ + vpandq _iHighMask(%rax), %zmm13, %zmm5 + +/* pH = yH*HH */ + vmulpd %zmm5, %zmm6, %zmm1 + +/* HL = T_Rh_Eh_HLLhi-HH */ + vsubpd %zmm5, %zmm13, %zmm4 + vpsrlq $32, %zmm1, %zmm14 + +/* pLL = y*HLL; + pHH = pH + *(double*)&db2p45_2p44 + */ + vaddpd _db2p45_2p44(%rax), %zmm1, %zmm10 + vpmovqd %zmm14, %ymm15 + vpandd _ABSMASK(%rax), %zmm15, %zmm14{%k1} + vpcmpd $5, _DOMAINRANGE(%rax), %zmm14, %k3{%k1} + +/* T1 = ((double*)exp2_tbl)[ 2*j ] */ + vpxord %zmm14, %zmm14, %zmm14 + vpbroadcastd %ecx, %zmm13{%k3}{z} + vpord %zmm13, %zmm11, %zmm11{%k1} + vptestmd %zmm11, %zmm11, %k0{%k1} + +/* pL=yL*HL+yH*HL; pL+=yL*HH */ + vmulpd %zmm4, %zmm2, %zmm11 + kmovw %k0, %ecx + vfmadd213pd %zmm11, %zmm4, %zmm6 + +/* pHH = pHH - *(double*)&db2p45_2p44 */ + vsubpd _db2p45_2p44(%rax), %zmm10, %zmm11 + vpmovqd %zmm10, %ymm4 + movzbl %cl, %ecx + +/* _n = Lo(pHH); + _n = _n & 0xffffff80; + _n = _n >> 7; + Hi(_2n) = (0x3ff+_n)<<20; Lo(_2n) = 0; -> 2^n + */ + vpslld $13, %zmm4, %zmm7{%k1} + +/* j = Lo(pHH)&0x0000007f */ + vpandd _jIndexMask(%rax), %zmm4, %zmm9{%k1} + vfmadd213pd %zmm6, %zmm5, %zmm2 + +/* pHL = pH - pHH */ + vsubpd %zmm11, %zmm1, %zmm1 + vpaddd _iOne(%rax), %zmm7, %zmm7{%k1} + +/* t=pL+pLL; t+=pHL */ + vfmadd231pd %zmm12, %zmm3, %zmm2 + vpslld $4, %zmm9, %zmm9{%k1} + kxnorw %k1, %k1, %k1 + vgatherdpd 36416(%rax,%ymm9), %zmm14{%k1} + vpmovzxdq %ymm7, %zmm8 + vaddpd %zmm1, %zmm2, %zmm2 + vmovups _cev_1(%rax), %zmm1 + vpsllq $32, %zmm8, %zmm13 + vpternlogq $168, _ifff0000000000000(%rax), %zmm8, %zmm13 + vfmadd213pd _cev_2(%rax), %zmm2, %zmm1 + vmulpd %zmm14, %zmm13, %zmm15 + vfmadd213pd _cev_3(%rax), %zmm2, %zmm1 + vmulpd %zmm2, %zmm15, %zmm3 + vfmadd213pd _cev_4(%rax), %zmm2, %zmm1 + vfmadd213pd _cev_5(%rax), %zmm2, %zmm1 + vfmadd213pd %zmm15, %zmm3, %zmm1 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + vmovaps %zmm1, %zmm0 + 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 + vmovups %zmm0, 1152(%rsp) + vmovups %zmm12, 1216(%rsp) + vmovups %zmm1, 1280(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + kmovw %k4, 1048(%rsp) + xorl %eax, %eax + kmovw %k5, 1040(%rsp) + kmovw %k6, 1032(%rsp) + kmovw %k7, 1024(%rsp) + vmovups %zmm16, 960(%rsp) + vmovups %zmm17, 896(%rsp) + vmovups %zmm18, 832(%rsp) + vmovups %zmm19, 768(%rsp) + vmovups %zmm20, 704(%rsp) + vmovups %zmm21, 640(%rsp) + vmovups %zmm22, 576(%rsp) + vmovups %zmm23, 512(%rsp) + vmovups %zmm24, 448(%rsp) + vmovups %zmm25, 384(%rsp) + vmovups %zmm26, 320(%rsp) + vmovups %zmm27, 256(%rsp) + vmovups %zmm28, 192(%rsp) + vmovups %zmm29, 128(%rsp) + vmovups %zmm30, 64(%rsp) + vmovups %zmm31, (%rsp) + movq %rsi, 1064(%rsp) + movq %rdi, 1056(%rsp) + movq %r12, 1096(%rsp) + cfi_offset_rel_rsp (12, 1096) + movb %dl, %r12b + movq %r13, 1088(%rsp) + cfi_offset_rel_rsp (13, 1088) + movl %ecx, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %eax, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + 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: + addb $1, %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_1_6 + + kmovw 1048(%rsp), %k4 + movq 1064(%rsp), %rsi + kmovw 1040(%rsp), %k5 + movq 1056(%rsp), %rdi + kmovw 1032(%rsp), %k6 + movq 1096(%rsp), %r12 + cfi_restore (%r12) + movq 1088(%rsp), %r13 + cfi_restore (%r13) + kmovw 1024(%rsp), %k7 + vmovups 960(%rsp), %zmm16 + vmovups 896(%rsp), %zmm17 + vmovups 832(%rsp), %zmm18 + vmovups 768(%rsp), %zmm19 + vmovups 704(%rsp), %zmm20 + vmovups 640(%rsp), %zmm21 + vmovups 576(%rsp), %zmm22 + vmovups 512(%rsp), %zmm23 + vmovups 448(%rsp), %zmm24 + vmovups 384(%rsp), %zmm25 + vmovups 320(%rsp), %zmm26 + vmovups 256(%rsp), %zmm27 + vmovups 192(%rsp), %zmm28 + vmovups 128(%rsp), %zmm29 + vmovups 64(%rsp), %zmm30 + vmovups (%rsp), %zmm31 + movq 1080(%rsp), %r14 + cfi_restore (%r14) + movq 1072(%rsp), %r15 + cfi_restore (%r15) + vmovups 1280(%rsp), %zmm1 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1160(%rsp,%r15), %xmm0 + vmovsd 1224(%rsp,%r15), %xmm1 + call pow@PLT + vmovsd %xmm0, 1288(%rsp,%r15) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1152(%rsp,%r15), %xmm0 + vmovsd 1216(%rsp,%r15), %xmm1 + call pow@PLT + vmovsd %xmm0, 1280(%rsp,%r15) + jmp .LBL_1_7 + +#endif +END (_ZGVeN8vv_pow_knl) + +ENTRY (_ZGVeN8vv_pow_skx) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512_ff _ZGVdN4vv_pow +#else + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1344, %rsp + vpsrlq $32, %zmm0, %zmm10 + kxnorw %k1, %k1, %k1 + kxnorw %k2, %k2, %k2 + kxnorw %k3, %k3, %k3 + vpmovqd %zmm10, %ymm7 + movq __svml_dpow_data@GOTPCREL(%rip), %rax + vmovaps %zmm1, %zmm6 + vpsrlq $32, %zmm6, %zmm13 + +/* i = (((Hi(x) & 0x000ffe00) + 0x00000200) >> 10); -> i = (b1..b11 + 1) / 2 */ + vpand _iIndexMask(%rax), %ymm7, %ymm15 + vpaddd _HIDELTA(%rax), %ymm7, %ymm2 + +/* k = Hi(x); k = k - 0x3fe7fe00; k = k >> 20 */ + vpsubd _i3fe7fe00(%rax), %ymm7, %ymm7 + vmovdqu _ABSMASK(%rax), %ymm4 + vmovdqu _LORANGE(%rax), %ymm3 + +/* x1 = x; Hi(x1) = (Hi(x1)&0x000fffff)|0x3ff00000 */ + vmovups _dbOne(%rax), %zmm11 + vmovdqu _INF(%rax), %ymm5 + vpaddd _iIndexAdd(%rax), %ymm15, %ymm12 + vpmovqd %zmm13, %ymm14 + vpternlogq $248, _iMantissaMask(%rax), %zmm0, %zmm11 + vpsrld $10, %ymm12, %ymm10 + vpsrad $20, %ymm7, %ymm13 + +/* Index for reciprocal table */ + vpslld $3, %ymm10, %ymm8 + +/* Index for log2 table */ + vpslld $4, %ymm10, %ymm1 + vpcmpgtd %ymm2, %ymm3, %ymm3 + vpand %ymm4, %ymm14, %ymm2 + vpaddd _i2p20_2p19(%rax), %ymm13, %ymm14 + vpmovzxdq %ymm14, %zmm15 + vpsllq $32, %zmm15, %zmm7 + vpternlogq $168, _iffffffff00000000(%rax), %zmm15, %zmm7 + vsubpd _db2p20_2p19(%rax), %zmm7, %zmm13 + vpxord %zmm9, %zmm9, %zmm9 + vgatherdpd 11712(%rax,%ymm8), %zmm9{%k1} + +/* T1 = ((double*)exp2_tbl)[ 2*j ] */ + kxnorw %k1, %k1, %k1 + vpxord %zmm12, %zmm12, %zmm12 + vpxord %zmm8, %zmm8, %zmm8 + vgatherdpd 19968(%rax,%ymm1), %zmm12{%k2} + vgatherdpd 19976(%rax,%ymm1), %zmm8{%k3} + vmovups _iHighMask(%rax), %zmm1 + +/* x1Hi=x1; Lo(x1Hi)&=0xf8000000; x1Lo = x1-x1Hi */ + vandpd %zmm1, %zmm11, %zmm10 + vsubpd %zmm10, %zmm11, %zmm15 + +/* r1 = x1*rcp1 */ + vmulpd %zmm9, %zmm11, %zmm11 + +/* E = -r1+__fence(x1Hi*rcp1) */ + vfmsub213pd %zmm11, %zmm9, %zmm10 + +/* cq = c+r1 */ + vaddpd _LHN(%rax), %zmm11, %zmm14 + +/* E=E+x1Lo*rcp1 */ + vfmadd213pd %zmm10, %zmm9, %zmm15 + +/* T = k + L1hi */ + vaddpd %zmm12, %zmm13, %zmm9 + +/* T_Rh = T + cq */ + vaddpd %zmm14, %zmm9, %zmm11 + +/* T_Rh_Eh = T_Rh + E */ + vaddpd %zmm15, %zmm11, %zmm13 + +/* Rl = T-T_Rh; -> -Rh */ + vsubpd %zmm11, %zmm9, %zmm12 + +/* HLL = T_Rh - T_Rh_Eh; -> -Eh */ + vsubpd %zmm13, %zmm11, %zmm9 + +/* Rl=Rl+cq */ + vaddpd %zmm12, %zmm14, %zmm10 + +/* HLL+=E; -> El */ + vaddpd %zmm9, %zmm15, %zmm7 + +/* HLL+=Rl */ + vaddpd %zmm10, %zmm7, %zmm12 + +/* 2^(y*(HH+HL+HLL)) starts here: + yH = y; Lo(yH)&=0xf8000000 + */ + vandpd %zmm1, %zmm6, %zmm7 + +/* HLL+=L1lo */ + vaddpd %zmm8, %zmm12, %zmm12 + +/* cq = cq + E */ + vaddpd %zmm15, %zmm14, %zmm8 + vmovups _clv_2(%rax), %zmm14 + +/* HLL = HLL + (((((((a7)*cq+a6)*cq+a5)*cq+a4)*cq+a3)*cq+a2)*cq+a1)*cq */ + vfmadd213pd _clv_3(%rax), %zmm8, %zmm14 + vfmadd213pd _clv_4(%rax), %zmm8, %zmm14 + vfmadd213pd _clv_5(%rax), %zmm8, %zmm14 + vfmadd213pd _clv_6(%rax), %zmm8, %zmm14 + vfmadd213pd _clv_7(%rax), %zmm8, %zmm14 + vfmadd213pd %zmm12, %zmm8, %zmm14 + +/* yL = y-yH */ + vsubpd %zmm7, %zmm6, %zmm8 + +/* T_Rh_Eh_HLLhi = T_Rh_Eh + HLL */ + vaddpd %zmm14, %zmm13, %zmm15 + +/* HH = T_Rh_Eh_HLLhi; Lo(HH)&=0xf8000000 */ + vandpd %zmm1, %zmm15, %zmm11 + +/* HLLhi = T_Rh_Eh_HLLhi - T_Rh_Eh */ + vsubpd %zmm13, %zmm15, %zmm13 + +/* pH = yH*HH */ + vmulpd %zmm11, %zmm7, %zmm9 + +/* HLL = HLL - HLLhi */ + vsubpd %zmm13, %zmm14, %zmm12 + +/* HL = T_Rh_Eh_HLLhi-HH */ + vsubpd %zmm11, %zmm15, %zmm10 + vpsrlq $32, %zmm9, %zmm1 + vmovdqu _DOMAINRANGE(%rax), %ymm13 + vpmovqd %zmm1, %ymm1 + vpand %ymm4, %ymm1, %ymm1 + vpcmpgtd %ymm5, %ymm2, %ymm4 + vpcmpeqd %ymm5, %ymm2, %ymm5 + vpternlogd $254, %ymm5, %ymm4, %ymm3 + vpcmpgtd %ymm13, %ymm1, %ymm2 + vpcmpeqd %ymm13, %ymm1, %ymm4 + vpternlogd $254, %ymm4, %ymm2, %ymm3 + +/* pLL = y*HLL */ + vmovups _db2p45_2p44(%rax), %zmm2 + +/* pHH = pH + *(double*)&db2p45_2p44 */ + vaddpd %zmm2, %zmm9, %zmm1 + vpmovqd %zmm1, %ymm5 + +/* j = Lo(pHH)&0x0000007f */ + vpand _jIndexMask(%rax), %ymm5, %ymm14 + vpslld $4, %ymm14, %ymm15 + vmovmskps %ymm3, %ecx + +/* pL=yL*HL+yH*HL; pL+=yL*HH */ + vmulpd %zmm10, %zmm8, %zmm3 + vfmadd213pd %zmm3, %zmm10, %zmm7 + vfmadd213pd %zmm7, %zmm11, %zmm8 + +/* _n = Lo(pHH) + _n = _n & 0xffffff80 + _n = _n >> 7 + Hi(_2n) = (0x3ff+_n)<<20; Lo(_2n) = 0; -> 2^n + */ + vpslld $13, %ymm5, %ymm7 + +/* t=pL+pLL; t+=pHL */ + vfmadd231pd %zmm6, %zmm12, %zmm8 + vpaddd _iOne(%rax), %ymm7, %ymm10 + vpmovzxdq %ymm10, %zmm11 + vpsllq $32, %zmm11, %zmm3 + vpternlogq $168, _ifff0000000000000(%rax), %zmm11, %zmm3 + +/* pHH = pHH - *(double*)&db2p45_2p44 */ + vsubpd %zmm2, %zmm1, %zmm11 + vmovups _cev_1(%rax), %zmm2 + +/* pHL = pH - pHH */ + vsubpd %zmm11, %zmm9, %zmm9 + vaddpd %zmm9, %zmm8, %zmm8 + vfmadd213pd _cev_2(%rax), %zmm8, %zmm2 + vfmadd213pd _cev_3(%rax), %zmm8, %zmm2 + vfmadd213pd _cev_4(%rax), %zmm8, %zmm2 + vfmadd213pd _cev_5(%rax), %zmm8, %zmm2 + vpxord %zmm4, %zmm4, %zmm4 + vgatherdpd 36416(%rax,%ymm15), %zmm4{%k1} + vmulpd %zmm4, %zmm3, %zmm1 + vmulpd %zmm8, %zmm1, %zmm12 + vfmadd213pd %zmm1, %zmm12, %zmm2 + testl %ecx, %ecx + jne .LBL_2_3 + +.LBL_2_2: + cfi_remember_state + vmovaps %zmm2, %zmm0 + movq %rbp, %rsp + cfi_def_cfa_register (%rsp) + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (%rbp) + ret + +.LBL_2_3: + cfi_restore_state + vmovups %zmm0, 1152(%rsp) + vmovups %zmm6, 1216(%rsp) + vmovups %zmm2, 1280(%rsp) + je .LBL_2_2 + + xorb %dl, %dl + xorl %eax, %eax + kmovw %k4, 1048(%rsp) + kmovw %k5, 1040(%rsp) + kmovw %k6, 1032(%rsp) + kmovw %k7, 1024(%rsp) + vmovups %zmm16, 960(%rsp) + vmovups %zmm17, 896(%rsp) + vmovups %zmm18, 832(%rsp) + vmovups %zmm19, 768(%rsp) + vmovups %zmm20, 704(%rsp) + vmovups %zmm21, 640(%rsp) + vmovups %zmm22, 576(%rsp) + vmovups %zmm23, 512(%rsp) + vmovups %zmm24, 448(%rsp) + vmovups %zmm25, 384(%rsp) + vmovups %zmm26, 320(%rsp) + vmovups %zmm27, 256(%rsp) + vmovups %zmm28, 192(%rsp) + vmovups %zmm29, 128(%rsp) + vmovups %zmm30, 64(%rsp) + vmovups %zmm31, (%rsp) + movq %rsi, 1064(%rsp) + movq %rdi, 1056(%rsp) + movq %r12, 1096(%rsp) + cfi_offset_rel_rsp (12, 1096) + movb %dl, %r12b + movq %r13, 1088(%rsp) + cfi_offset_rel_rsp (13, 1088) + movl %ecx, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %eax, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + cfi_remember_state + +.LBL_2_6: + btl %r14d, %r13d + jc .LBL_2_12 + +.LBL_2_7: + lea 1(%r14), %esi + btl %esi, %r13d + jc .LBL_2_10 + +.LBL_2_8: + incb %r12b + addl $2, %r14d + cmpb $16, %r12b + jb .LBL_2_6 + + kmovw 1048(%rsp), %k4 + kmovw 1040(%rsp), %k5 + kmovw 1032(%rsp), %k6 + kmovw 1024(%rsp), %k7 + vmovups 960(%rsp), %zmm16 + vmovups 896(%rsp), %zmm17 + vmovups 832(%rsp), %zmm18 + vmovups 768(%rsp), %zmm19 + vmovups 704(%rsp), %zmm20 + vmovups 640(%rsp), %zmm21 + vmovups 576(%rsp), %zmm22 + vmovups 512(%rsp), %zmm23 + vmovups 448(%rsp), %zmm24 + vmovups 384(%rsp), %zmm25 + vmovups 320(%rsp), %zmm26 + vmovups 256(%rsp), %zmm27 + vmovups 192(%rsp), %zmm28 + vmovups 128(%rsp), %zmm29 + vmovups 64(%rsp), %zmm30 + vmovups (%rsp), %zmm31 + vmovups 1280(%rsp), %zmm2 + movq 1064(%rsp), %rsi + movq 1056(%rsp), %rdi + movq 1096(%rsp), %r12 + cfi_restore (%r12) + movq 1088(%rsp), %r13 + cfi_restore (%r13) + movq 1080(%rsp), %r14 + cfi_restore (%r14) + movq 1072(%rsp), %r15 + cfi_restore (%r15) + jmp .LBL_2_2 + +.LBL_2_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1224(%rsp,%r15), %xmm1 + vzeroupper + vmovsd 1160(%rsp,%r15), %xmm0 + + call pow@PLT + + vmovsd %xmm0, 1288(%rsp,%r15) + jmp .LBL_2_8 + +.LBL_2_12: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1216(%rsp,%r15), %xmm1 + vzeroupper + vmovsd 1152(%rsp,%r15), %xmm0 + + call pow@PLT + + vmovsd %xmm0, 1280(%rsp,%r15) + jmp .LBL_2_7 + +#endif +END (_ZGVeN8vv_pow_skx) |