diff options
Diffstat (limited to 'sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S')
-rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S | 593 |
1 files changed, 593 insertions, 0 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S new file mode 100644 index 0000000000..fcbf3935f1 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S @@ -0,0 +1,593 @@ +/* Function sincos 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_sincos_data.h" +#include "svml_d_wrapper_impl.h" + +/* + ALGORITHM DESCRIPTION: + + ( low accuracy ( < 4ulp ) or enhanced performance + ( half of correct mantissa ) implementation ) + + Argument representation: + arg = N*Pi + R + + Result calculation: + sin(arg) = sin(N*Pi + R) = (-1)^N * sin(R) + arg + Pi/2 = (N'*Pi + R') + cos(arg) = sin(arg+Pi/2) = sin(N'*Pi + R') = (-1)^N' * sin(R') + sin(R), sin(R') are approximated by corresponding polynomial. */ + + .text +ENTRY (_ZGVeN8vvv_sincos_knl) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512_fFF _ZGVdN4vvv_sincos +#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 + movq __svml_dsincos_data@GOTPCREL(%rip), %rax + vmovaps %zmm0, %zmm4 + movq $-1, %rdx + vmovups __dSignMask(%rax), %zmm12 + vmovups __dInvPI(%rax), %zmm5 + +/* ARGUMENT RANGE REDUCTION: + Absolute argument: X' = |X| */ + vpandnq %zmm4, %zmm12, %zmm3 + vmovups __dPI1_FMA(%rax), %zmm7 + vmovups __dPI3_FMA(%rax), %zmm9 + +/* SinR = X' - SinN*Pi1 */ + vmovaps %zmm3, %zmm8 + +/* CosR = SinX - CosN*Pi1 */ + vmovaps %zmm3, %zmm10 + +/* SinY = X'*InvPi + RS : right shifter add */ + vfmadd213pd __dRShifter(%rax), %zmm3, %zmm5 + vmovups __dC6(%rax), %zmm13 + +/* SinN = Y - RS : right shifter sub */ + vsubpd __dRShifter(%rax), %zmm5, %zmm1 + vmovaps %zmm13, %zmm14 + +/* SinSignRes = Y<<63 : shift LSB to MSB place for result sign */ + vpsllq $63, %zmm5, %zmm2 + vcmppd $22, __dRangeVal(%rax), %zmm3, %k1 + +/* Update CosRSign and CosSignRes signs */ + vmovaps %zmm12, %zmm5 + vfnmadd231pd %zmm1, %zmm7, %zmm8 + +/* SinR = SinR - SinN*Pi1 */ + vfnmadd231pd __dPI2_FMA(%rax), %zmm1, %zmm8 + +/* Sine result sign: SinRSign = SignMask & SinR */ + vpandq %zmm8, %zmm12, %zmm11 + +/* Set SinRSign to 0.5 */ + vporq __dOneHalf(%rax), %zmm11, %zmm6 + vpternlogq $150, %zmm2, %zmm11, %zmm5 + +/* Update sign SinSignRes */ + vpternlogq $120, %zmm4, %zmm12, %zmm2 + +/* Polynomial approximation */ + vmovups __dC7(%rax), %zmm11 + +/* CosN = SinN +(-)0.5 */ + vaddpd %zmm6, %zmm1, %zmm0 + +/* SinR = SinR - SinN*Pi3 */ + vfnmadd213pd %zmm8, %zmm9, %zmm1 + vfnmadd231pd %zmm0, %zmm7, %zmm10 + +/* SinR2 = SinR^2 */ + vmulpd %zmm1, %zmm1, %zmm15 + +/* Grab SignX + CosR = CosR - CosN*Pi2 */ + vfnmadd231pd __dPI2_FMA(%rax), %zmm0, %zmm10 + vfmadd231pd __dC7(%rax), %zmm15, %zmm14 + +/* CosR = CosR - CosN*Pi3 */ + vfnmadd213pd %zmm10, %zmm9, %zmm0 + vfmadd213pd __dC5(%rax), %zmm15, %zmm14 + +/* CosR2 = CosR^2 */ + vmulpd %zmm0, %zmm0, %zmm12 + vfmadd213pd __dC4(%rax), %zmm15, %zmm14 + vfmadd213pd %zmm13, %zmm12, %zmm11 + +/* SinPoly = C3 + SinR2*(C4 + SinR2*(C5 + SinR2*(C6 + SinR2*C7))) */ + vfmadd213pd __dC3(%rax), %zmm15, %zmm14 + vfmadd213pd __dC5(%rax), %zmm12, %zmm11 + +/* SinPoly = C2 + SinR2*SinPoly */ + vfmadd213pd __dC2(%rax), %zmm15, %zmm14 + vfmadd213pd __dC4(%rax), %zmm12, %zmm11 + +/* SinPoly = C1 + SinR2*SinPoly */ + vfmadd213pd __dC1(%rax), %zmm15, %zmm14 + +/* CosPoly = C3 + CosR2*(C4 + CosR2*(C5 + CosR2*(C6 + CosR2*C7))) */ + vfmadd213pd __dC3(%rax), %zmm12, %zmm11 + +/* SinPoly = SinR2*SinPoly */ + vmulpd %zmm15, %zmm14, %zmm13 + +/* CosPoly = C2 + CosR2*CosPoly */ + vfmadd213pd __dC2(%rax), %zmm12, %zmm11 + +/* SinPoly = SinR*SinPoly */ + vfmadd213pd %zmm1, %zmm1, %zmm13 + vpbroadcastq %rdx, %zmm1{%k1}{z} + +/* CosPoly = C1 + CosR2*CosPoly */ + vfmadd213pd __dC1(%rax), %zmm12, %zmm11 + vptestmq %zmm1, %zmm1, %k0 + kmovw %k0, %ecx + +/* CosPoly = CosR2*CosPoly */ + vmulpd %zmm12, %zmm11, %zmm14 + movzbl %cl, %ecx + +/* CosPoly = CosR*CosPoly */ + vfmadd213pd %zmm0, %zmm0, %zmm14 + +/* Final reconstruction. + Update Sin result's sign */ + vpxorq %zmm2, %zmm13, %zmm0 + +/* Update Cos result's sign */ + vpxorq %zmm5, %zmm14, %zmm2 + testl %ecx, %ecx + jne .LBL_1_3 + +.LBL_1_2: + cfi_remember_state + vmovups %zmm0, (%rdi) + vmovups %zmm2, (%rsi) + 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 %zmm4, 1152(%rsp) + vmovups %zmm0, 1216(%rsp) + vmovups %zmm2, 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, 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 %eax, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %ecx, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + movq %rbx, 1064(%rsp) + movq %rdi, %rbx + cfi_remember_state + +.LBL_1_6: + btl %r13d, %r14d + jc .LBL_1_13 + +.LBL_1_7: + lea 1(%r13), %esi + btl %esi, %r14d + jc .LBL_1_10 + +.LBL_1_8: + addb $1, %r12b + addl $2, %r13d + cmpb $16, %r12b + jb .LBL_1_6 + + movq %rbx, %rdi + kmovw 1048(%rsp), %k4 + movq 1056(%rsp), %rsi + kmovw 1040(%rsp), %k5 + movq 1096(%rsp), %r12 + cfi_restore (%r12) + kmovw 1032(%rsp), %k6 + 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) + movq 1064(%rsp), %rbx + vmovups 1216(%rsp), %zmm0 + vmovups 1280(%rsp), %zmm2 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1160(%rsp,%r15), %xmm0 + + call sin@PLT + + vmovsd %xmm0, 1224(%rsp,%r15) + vmovsd 1160(%rsp,%r15), %xmm0 + + call cos@PLT + + vmovsd %xmm0, 1288(%rsp,%r15) + jmp .LBL_1_8 + +.LBL_1_13: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1152(%rsp,%r15), %xmm0 + + call sin@PLT + + vmovsd %xmm0, 1216(%rsp,%r15) + vmovsd 1152(%rsp,%r15), %xmm0 + + call cos@PLT + + vmovsd %xmm0, 1280(%rsp,%r15) + jmp .LBL_1_7 + +#endif +END (_ZGVeN8vvv_sincos_knl) + +ENTRY (_ZGVeN8vvv_sincos_skx) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512_fFF _ZGVdN4vvv_sincos +#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 + movq __svml_dsincos_data@GOTPCREL(%rip), %rax + vmovaps %zmm0, %zmm8 + vmovups __dSignMask(%rax), %zmm4 + vmovups __dInvPI(%rax), %zmm9 + vmovups __dRShifter(%rax), %zmm10 + vmovups __dPI1_FMA(%rax), %zmm13 + vmovups __dPI2_FMA(%rax), %zmm14 + vmovups __dOneHalf(%rax), %zmm11 + vmovups __dPI3_FMA(%rax), %zmm2 + +/* ARGUMENT RANGE REDUCTION: + Absolute argument: X' = |X| */ + vandnpd %zmm8, %zmm4, %zmm7 + +/* SinY = X'*InvPi + RS : right shifter add */ + vfmadd213pd %zmm10, %zmm7, %zmm9 + vcmppd $18, __dRangeVal(%rax), %zmm7, %k1 + +/* SinSignRes = Y<<63 : shift LSB to MSB place for result sign */ + vpsllq $63, %zmm9, %zmm6 + +/* SinN = Y - RS : right shifter sub */ + vsubpd %zmm10, %zmm9, %zmm5 + vmovups __dC5(%rax), %zmm9 + vmovups __dC4(%rax), %zmm10 + +/* SinR = X' - SinN*Pi1 */ + vmovaps %zmm7, %zmm15 + vfnmadd231pd %zmm5, %zmm13, %zmm15 + +/* SinR = SinR - SinN*Pi1 */ + vfnmadd231pd %zmm5, %zmm14, %zmm15 + +/* Sine result sign: SinRSign = SignMask & SinR */ + vandpd %zmm15, %zmm4, %zmm1 + +/* Set SinRSign to 0.5 */ + vorpd %zmm1, %zmm11, %zmm12 + vmovups __dC3(%rax), %zmm11 + +/* CosN = SinN +(-)0.5 */ + vaddpd %zmm12, %zmm5, %zmm3 + +/* SinR = SinR - SinN*Pi3 */ + vfnmadd213pd %zmm15, %zmm2, %zmm5 + vmovups __dC2(%rax), %zmm12 + +/* SinR2 = SinR^2 */ + vmulpd %zmm5, %zmm5, %zmm15 + +/* CosR = SinX - CosN*Pi1 */ + vmovaps %zmm7, %zmm0 + vfnmadd231pd %zmm3, %zmm13, %zmm0 + vmovups __dC1(%rax), %zmm13 + +/* Grab SignX + CosR = CosR - CosN*Pi2 */ + vfnmadd231pd %zmm3, %zmm14, %zmm0 + +/* CosR = CosR - CosN*Pi3 */ + vfnmadd213pd %zmm0, %zmm2, %zmm3 + +/* Polynomial approximation */ + vmovups __dC7(%rax), %zmm0 + +/* Update CosRSign and CosSignRes signs */ + vmovaps %zmm4, %zmm2 + vpternlogq $150, %zmm6, %zmm1, %zmm2 + +/* Update sign SinSignRes */ + vpternlogq $120, %zmm8, %zmm4, %zmm6 + +/* CosR2 = CosR^2 */ + vmulpd %zmm3, %zmm3, %zmm1 + vmovups __dC6(%rax), %zmm4 + vmovaps %zmm0, %zmm14 + vfmadd213pd %zmm4, %zmm1, %zmm0 + vfmadd213pd %zmm4, %zmm15, %zmm14 + vfmadd213pd %zmm9, %zmm1, %zmm0 + vfmadd213pd %zmm9, %zmm15, %zmm14 + vfmadd213pd %zmm10, %zmm1, %zmm0 + vfmadd213pd %zmm10, %zmm15, %zmm14 + +/* CosPoly = C3 + CosR2*(C4 + CosR2*(C5 + CosR2*(C6 + CosR2*C7))) */ + vfmadd213pd %zmm11, %zmm1, %zmm0 + +/* SinPoly = C3 + SinR2*(C4 + SinR2*(C5 + SinR2*(C6 + SinR2*C7))) */ + vfmadd213pd %zmm11, %zmm15, %zmm14 + +/* CosPoly = C2 + CosR2*CosPoly */ + vfmadd213pd %zmm12, %zmm1, %zmm0 + +/* SinPoly = C2 + SinR2*SinPoly */ + vfmadd213pd %zmm12, %zmm15, %zmm14 + +/* CosPoly = C1 + CosR2*CosPoly */ + vfmadd213pd %zmm13, %zmm1, %zmm0 + +/* SinPoly = C1 + SinR2*SinPoly */ + vfmadd213pd %zmm13, %zmm15, %zmm14 + +/* CosPoly = CosR2*CosPoly */ + vmulpd %zmm1, %zmm0, %zmm1 + +/* SinPoly = SinR2*SinPoly */ + vmulpd %zmm15, %zmm14, %zmm4 + +/* CosPoly = CosR*CosPoly */ + vfmadd213pd %zmm3, %zmm3, %zmm1 + +/* SinPoly = SinR*SinPoly */ + vfmadd213pd %zmm5, %zmm5, %zmm4 + vpbroadcastq .L_2il0floatpacket.15(%rip), %zmm3 + +/* Update Cos result's sign */ + vxorpd %zmm2, %zmm1, %zmm1 + +/* Final reconstruction. + Update Sin result's sign */ + vxorpd %zmm6, %zmm4, %zmm0 + vpandnq %zmm7, %zmm7, %zmm3{%k1} + vcmppd $3, %zmm3, %zmm3, %k0 + kmovw %k0, %ecx + testl %ecx, %ecx + jne .LBL_2_3 + +.LBL_2_2: + cfi_remember_state + vmovups %zmm0, (%rdi) + vmovups %zmm1, (%rsi) + 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 %zmm8, 1152(%rsp) + vmovups %zmm0, 1216(%rsp) + vmovups %zmm1, 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, 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 %eax, %r13d + movq %r14, 1080(%rsp) + cfi_offset_rel_rsp (14, 1080) + movl %ecx, %r14d + movq %r15, 1072(%rsp) + cfi_offset_rel_rsp (15, 1072) + movq %rbx, 1064(%rsp) + movq %rdi, %rbx + cfi_remember_state + +.LBL_2_6: + btl %r13d, %r14d + jc .LBL_2_13 + +.LBL_2_7: + lea 1(%r13), %esi + btl %esi, %r14d + jc .LBL_2_10 + +.LBL_2_8: + incb %r12b + addl $2, %r13d + cmpb $16, %r12b + jb .LBL_2_6 + + kmovw 1048(%rsp), %k4 + movq %rbx, %rdi + 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 + vmovups 1280(%rsp), %zmm1 + movq 1056(%rsp), %rsi + 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) + movq 1064(%rsp), %rbx + 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 sin@PLT + + vmovsd %xmm0, 1224(%rsp,%r15) + vmovsd 1160(%rsp,%r15), %xmm0 + + call cos@PLT + + vmovsd %xmm0, 1288(%rsp,%r15) + jmp .LBL_2_8 + +.LBL_2_13: + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1152(%rsp,%r15), %xmm0 + vzeroupper + vmovsd 1152(%rsp,%r15), %xmm0 + + call sin@PLT + + vmovsd %xmm0, 1216(%rsp,%r15) + vmovsd 1152(%rsp,%r15), %xmm0 + + call cos@PLT + + vmovsd %xmm0, 1280(%rsp,%r15) + jmp .LBL_2_7 + +#endif +END (_ZGVeN8vvv_sincos_skx) + + .section .rodata, "a" +.L_2il0floatpacket.15: + .long 0xffffffff,0xffffffff + .type .L_2il0floatpacket.15,@object |