diff options
Diffstat (limited to 'sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S')
| -rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S | 468 |
1 files changed, 468 insertions, 0 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S new file mode 100644 index 0000000000..b0f3dd580c --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S @@ -0,0 +1,468 @@ +/* Function log 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_log_data.h" +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVeN8v_log_knl) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN4v_log +#else +/* + ALGORITHM DESCRIPTION: + + log(x) = -log(Rcp) + log(Rcp*x), + where Rcp ~ 1/x (accuracy ~9 bits, obtained by + rounding HW approximation to 1+9 mantissa bits) + + Reduced argument R=Rcp*x-1 is used to approximate log(1+R) as polynomial + + log(Rcp) = exponent_Rcp*log(2) + log(mantissa_Rcp) + -log(mantissa_Rcp) is obtained from a lookup table, + accessed by a 9-bit index + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1280, %rsp + movq __svml_dlog_data@GOTPCREL(%rip), %rdx + movq $-1, %rax + +/* isolate exponent bits */ + vpsrlq $20, %zmm0, %zmm2 + vpsrlq $32, %zmm2, %zmm3 + vpxord %zmm2, %zmm2, %zmm2 + kxnorw %k3, %k3, %k3 + vmovups _Two10(%rdx), %zmm1 + vmovups _One(%rdx), %zmm9 + vpmovqd %zmm3, %ymm4 + +/* convert biased exponent to DP format */ + vcvtdq2pd %ymm4, %zmm13 + +/* preserve mantissa, set input exponent to 2^(-10) */ + vpternlogq $248, _ExpMask(%rdx), %zmm0, %zmm1 + vcmppd $17, _MinNorm(%rdx), %zmm0, %k1 + +/* reciprocal approximation good to at least 11 bits */ + vrcp28pd %zmm1, %zmm5 + vpbroadcastq %rax, %zmm6{%k1}{z} + vmovups _poly_coeff_3(%rdx), %zmm15 + vcmppd $22, _MaxNorm(%rdx), %zmm0, %k2 + vmovups _Bias1(%rdx), %zmm14 + +/* round reciprocal to nearest integer, will have 1+9 mantissa bits */ + vrndscalepd $8, %zmm5, %zmm11 + vpbroadcastq %rax, %zmm7{%k2}{z} + +/* argument reduction started: R = Mantissa*Rcp - 1 */ + vfmsub213pd %zmm9, %zmm11, %zmm1 + +/* calculate index for table lookup */ + vpsrlq $40, %zmm11, %zmm10 + vgatherqpd _LogRcp_lookup(%rdx,%zmm10), %zmm2{%k3} + vcmppd $30, _Threshold(%rdx), %zmm11, %k1 + +/* combine and get argument value range mask */ + vporq %zmm7, %zmm6, %zmm8 + +/* exponent*log(2.0) */ + vmovups _poly_coeff_1(%rdx), %zmm11 + vmulpd %zmm1, %zmm1, %zmm10 + vptestmq %zmm8, %zmm8, %k0 + vfmadd213pd _poly_coeff_4(%rdx), %zmm1, %zmm15 + kmovw %k0, %ecx + +/* polynomial computation */ + vfmadd213pd _poly_coeff_2(%rdx), %zmm1, %zmm11 + movzbl %cl, %ecx + vpbroadcastq %rax, %zmm12{%k1}{z} + vfmadd213pd %zmm15, %zmm10, %zmm11 + vpternlogq $248, _Bias(%rdx), %zmm12, %zmm14 + +/* + Table stores -log(0.5*mantissa) for larger mantissas, + adjust exponent accordingly + */ + vsubpd %zmm14, %zmm13, %zmm3 + +/* + reconstruction: + (exponent*log(2)) + (LogRcp + (R+poly)) + */ + vfmadd213pd %zmm1, %zmm10, %zmm11 + vaddpd %zmm2, %zmm11, %zmm1 + vfmadd132pd _L2(%rdx), %zmm1, %zmm3 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + vmovaps %zmm3, %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 %zmm3, 1216(%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 1216(%rsp), %zmm3 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1160(%rsp,%r15), %xmm0 + call log@PLT + vmovsd %xmm0, 1224(%rsp,%r15) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1152(%rsp,%r15), %xmm0 + call log@PLT + vmovsd %xmm0, 1216(%rsp,%r15) + jmp .LBL_1_7 +#endif +END (_ZGVeN8v_log_knl) + +ENTRY (_ZGVeN8v_log_skx) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN4v_log +#else +/* + ALGORITHM DESCRIPTION: + + log(x) = -log(Rcp) + log(Rcp*x), + where Rcp ~ 1/x (accuracy ~9 bits, + obtained by rounding HW approximation to 1+9 mantissa bits) + + Reduced argument R=Rcp*x-1 is used to approximate log(1+R) as polynomial + + log(Rcp) = exponent_Rcp*log(2) + log(mantissa_Rcp) + -log(mantissa_Rcp) is obtained from a lookup table, + accessed by a 9-bit index + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $1280, %rsp + movq __svml_dlog_data@GOTPCREL(%rip), %rax + vmovaps %zmm0, %zmm3 + kxnorw %k3, %k3, %k3 + vmovups _Two10(%rax), %zmm2 + vmovups _Threshold(%rax), %zmm14 + vmovups _One(%rax), %zmm11 + vcmppd $21, _MinNorm(%rax), %zmm3, %k1 + vcmppd $18, _MaxNorm(%rax), %zmm3, %k2 + +/* isolate exponent bits */ + vpsrlq $20, %zmm3, %zmm4 + +/* preserve mantissa, set input exponent to 2^(-10) */ + vpternlogq $248, _ExpMask(%rax), %zmm3, %zmm2 + vpbroadcastq .L_2il0floatpacket.12(%rip), %zmm1 + vpsrlq $32, %zmm4, %zmm6 + +/* reciprocal approximation good to at least 11 bits */ + vrcp14pd %zmm2, %zmm5 + +/* exponent*log(2.0) */ + vmovups _poly_coeff_1(%rax), %zmm4 + vpmovqd %zmm6, %ymm7 + +/* round reciprocal to nearest integer, will have 1+9 mantissa bits */ + vrndscalepd $8, %zmm5, %zmm0 + +/* calculate index for table lookup */ + vpsrlq $40, %zmm0, %zmm12 + +/* argument reduction started: R = Mantissa*Rcp - 1 */ + vfmsub213pd %zmm11, %zmm0, %zmm2 + vpmovqd %zmm12, %ymm13 + +/* polynomial computation */ + vfmadd213pd _poly_coeff_2(%rax), %zmm2, %zmm4 + vmovaps %zmm1, %zmm8 + vmovaps %zmm1, %zmm9 + vpxord %zmm5, %zmm5, %zmm5 + vgatherdpd _LogRcp_lookup(%rax,%ymm13), %zmm5{%k3} + vmovups _Bias1(%rax), %zmm13 + vpandnq %zmm3, %zmm3, %zmm8{%k1} + vcmppd $21, %zmm0, %zmm14, %k1 + vpandnq %zmm14, %zmm14, %zmm1{%k1} + vmulpd %zmm2, %zmm2, %zmm14 + vpternlogq $248, _Bias(%rax), %zmm1, %zmm13 + vmovups _poly_coeff_3(%rax), %zmm1 + vfmadd213pd _poly_coeff_4(%rax), %zmm2, %zmm1 + vfmadd213pd %zmm1, %zmm14, %zmm4 + +/* + reconstruction: + (exponent*log(2)) + (LogRcp + (R+poly)) + */ + vfmadd213pd %zmm2, %zmm14, %zmm4 + vaddpd %zmm5, %zmm4, %zmm2 + vpandnq %zmm3, %zmm3, %zmm9{%k2} + +/* combine and get argument value range mask */ + vorpd %zmm9, %zmm8, %zmm10 + vcmppd $3, %zmm10, %zmm10, %k0 + kmovw %k0, %ecx + +/* convert biased exponent to DP format */ + vcvtdq2pd %ymm7, %zmm15 + +/* + Table stores -log(0.5*mantissa) for larger mantissas, + adjust exponent accordingly + */ + vsubpd %zmm13, %zmm15, %zmm0 + vfmadd132pd _L2(%rax), %zmm2, %zmm0 + testl %ecx, %ecx + jne .LBL_2_3 + +.LBL_2_2: + cfi_remember_state + 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 %zmm3, 1152(%rsp) + vmovups %zmm0, 1216(%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 1216(%rsp), %zmm0 + 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 1160(%rsp,%r15), %xmm0 + vzeroupper + vmovsd 1160(%rsp,%r15), %xmm0 + + call log@PLT + + vmovsd %xmm0, 1224(%rsp,%r15) + jmp .LBL_2_8 + +.LBL_2_12: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1152(%rsp,%r15), %xmm0 + vzeroupper + vmovsd 1152(%rsp,%r15), %xmm0 + + call log@PLT + + vmovsd %xmm0, 1216(%rsp,%r15) + jmp .LBL_2_7 +#endif +END (_ZGVeN8v_log_skx) + + .section .rodata, "a" +.L_2il0floatpacket.12: + .long 0xffffffff,0xffffffff + .type .L_2il0floatpacket.12,@object |