From 2193311288b97cf11dfabf1be22eac89b4ff7366 Mon Sep 17 00:00:00 2001 From: Andrew Senkevich Date: Tue, 9 Jun 2015 14:25:49 +0300 Subject: Start of series of patches with x86_64 vector math functions. Here is implementation of cos containing SSE, AVX, AVX2 and AVX512 versions according to Vector ABI which had been discussed in . Vector math library build and ABI testing enabled by default for x86_64. * sysdeps/x86_64/fpu/Makefile: New file. * sysdeps/x86_64/fpu/Versions: New file. * sysdeps/x86_64/fpu/svml_d_cos_data.S: New file. * sysdeps/x86_64/fpu/svml_d_cos_data.h: New file. * sysdeps/x86_64/fpu/svml_d_cos2_core.S: New file. * sysdeps/x86_64/fpu/svml_d_cos4_core.S: New file. * sysdeps/x86_64/fpu/svml_d_cos4_core_avx.S: New file. * sysdeps/x86_64/fpu/svml_d_cos8_core.S: New file. * sysdeps/x86_64/fpu/svml_d_wrapper_impl.h: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S: New file. * sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added build of SSE, AVX2 and AVX512 IFUNC versions. * sysdeps/x86/fpu/bits/math-vector.h: Added SIMD declaration for cos. * math/bits/mathcalls.h: Added cos declaration with __MATHCALL_VEC. * sysdeps/x86_64/configure.ac: Options for libmvec build. * sysdeps/x86_64/configure: Regenerated. * sysdeps/x86_64/sysdep.h (cfi_offset_rel_rsp): New macro. * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New file. * manual/install.texi (Configuring and compiling): Document --disable-mathvec. * INSTALL: Regenerated. * NEWS: Mention addition of libmvec and x86_64 vector cos. --- sysdeps/x86_64/fpu/multiarch/Makefile | 5 + sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S | 38 ++ .../x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S | 223 ++++++++++ sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S | 38 ++ .../x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S | 207 +++++++++ sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S | 39 ++ .../x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S | 463 +++++++++++++++++++++ 7 files changed, 1013 insertions(+) create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S create mode 100644 sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S (limited to 'sysdeps/x86_64/fpu/multiarch') diff --git a/sysdeps/x86_64/fpu/multiarch/Makefile b/sysdeps/x86_64/fpu/multiarch/Makefile index 12b0526..b2f3266 100644 --- a/sysdeps/x86_64/fpu/multiarch/Makefile +++ b/sysdeps/x86_64/fpu/multiarch/Makefile @@ -51,3 +51,8 @@ CFLAGS-slowexp-avx.c = -msse2avx -DSSE2AVX CFLAGS-s_tan-avx.c = -msse2avx -DSSE2AVX endif endif + +ifeq ($(subdir),mathvec) +libmvec-sysdep_routines += svml_d_cos2_core_sse4 svml_d_cos4_core_avx2 \ + svml_d_cos8_core_avx512 +endif diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S new file mode 100644 index 0000000..5f67d83 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core.S @@ -0,0 +1,38 @@ +/* Multiple versions of vectorized cos, vector length is 2. + 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 + . */ + +#include +#include + + .text +ENTRY (_ZGVbN2v_cos) + .type _ZGVbN2v_cos, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1f + call __init_cpu_features +1: leaq _ZGVbN2v_cos_sse4(%rip), %rax + testl $bit_SSE4_1, __cpu_features+CPUID_OFFSET+index_SSE4_1(%rip) + jz 2f + ret +2: leaq _ZGVbN2v_cos_sse2(%rip), %rax + ret +END (_ZGVbN2v_cos) +libmvec_hidden_def (_ZGVbN2v_cos) + +#define _ZGVbN2v_cos _ZGVbN2v_cos_sse2 +#include "../svml_d_cos2_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S new file mode 100644 index 0000000..11348a3 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos2_core_sse4.S @@ -0,0 +1,223 @@ +/* Function cos vectorized with SSE4. + 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 + . */ + +#include +#include "svml_d_cos_data.h" + + .text +ENTRY (_ZGVbN2v_cos_sse4) +/* ALGORITHM DESCRIPTION: + + ( low accuracy ( < 4ulp ) or enhanced performance + ( half of correct mantissa ) implementation ) + + Argument representation: + arg + Pi/2 = (N*Pi + R) + + Result calculation: + cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R) + sin(R) is approximated by corresponding polynomial + */ + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_rel_offset (%rbp, 0) + movq %rsp, %rbp + cfi_def_cfa_register (%rbp) + andq $-64, %rsp + subq $320, %rsp + movaps %xmm0, %xmm3 + movq __svml_dcos_data@GOTPCREL(%rip), %rax + movups __dHalfPI(%rax), %xmm2 + +/* ARGUMENT RANGE REDUCTION: + Add Pi/2 to argument: X' = X+Pi/2 + */ + addpd %xmm3, %xmm2 + movups __dInvPI(%rax), %xmm5 + movups __dAbsMask(%rax), %xmm4 + +/* Get absolute argument value: X' = |X'| */ + andps %xmm2, %xmm4 + +/* Y = X'*InvPi + RS : right shifter add */ + mulpd %xmm5, %xmm2 + +/* Check for large arguments path */ + cmpnlepd __dRangeVal(%rax), %xmm4 + movups __dRShifter(%rax), %xmm6 + addpd %xmm6, %xmm2 + movmskpd %xmm4, %ecx + +/* N = Y - RS : right shifter sub */ + movaps %xmm2, %xmm1 + +/* SignRes = Y<<63 : shift LSB to MSB place for result sign */ + psllq $63, %xmm2 + subpd %xmm6, %xmm1 + +/* N = N - 0.5 */ + subpd __dOneHalf(%rax), %xmm1 + movups __dPI1(%rax), %xmm7 + +/* R = X - N*Pi1 */ + mulpd %xmm1, %xmm7 + movups __dPI2(%rax), %xmm4 + +/* R = R - N*Pi2 */ + mulpd %xmm1, %xmm4 + subpd %xmm7, %xmm0 + movups __dPI3(%rax), %xmm5 + +/* R = R - N*Pi3 */ + mulpd %xmm1, %xmm5 + subpd %xmm4, %xmm0 + +/* R = R - N*Pi4 */ + movups __dPI4(%rax), %xmm6 + mulpd %xmm6, %xmm1 + subpd %xmm5, %xmm0 + subpd %xmm1, %xmm0 + +/* POLYNOMIAL APPROXIMATION: R2 = R*R */ + movaps %xmm0, %xmm4 + mulpd %xmm0, %xmm4 + movups __dC7(%rax), %xmm1 + mulpd %xmm4, %xmm1 + addpd __dC6(%rax), %xmm1 + mulpd %xmm4, %xmm1 + addpd __dC5(%rax), %xmm1 + mulpd %xmm4, %xmm1 + addpd __dC4(%rax), %xmm1 + +/* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */ + mulpd %xmm4, %xmm1 + addpd __dC3(%rax), %xmm1 + +/* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */ + mulpd %xmm4, %xmm1 + addpd __dC2(%rax), %xmm1 + mulpd %xmm4, %xmm1 + addpd __dC1(%rax), %xmm1 + mulpd %xmm1, %xmm4 + mulpd %xmm0, %xmm4 + addpd %xmm4, %xmm0 + +/* RECONSTRUCTION: + Final sign setting: Res = Poly^SignRes */ + xorps %xmm2, %xmm0 + 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 + movups %xmm3, 192(%rsp) + movups %xmm0, 256(%rsp) + je .LBL_1_2 + + xorb %dl, %dl + xorl %eax, %eax + movups %xmm8, 112(%rsp) + movups %xmm9, 96(%rsp) + movups %xmm10, 80(%rsp) + movups %xmm11, 64(%rsp) + movups %xmm12, 48(%rsp) + movups %xmm13, 32(%rsp) + movups %xmm14, 16(%rsp) + movups %xmm15, (%rsp) + movq %rsi, 136(%rsp) + movq %rdi, 128(%rsp) + movq %r12, 168(%rsp) + cfi_offset_rel_rsp (12, 168) + movb %dl, %r12b + movq %r13, 160(%rsp) + cfi_offset_rel_rsp (13, 160) + movl %ecx, %r13d + movq %r14, 152(%rsp) + cfi_offset_rel_rsp (14, 152) + movl %eax, %r14d + movq %r15, 144(%rsp) + cfi_offset_rel_rsp (15, 144) + 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 + + movups 112(%rsp), %xmm8 + movups 96(%rsp), %xmm9 + movups 80(%rsp), %xmm10 + movups 64(%rsp), %xmm11 + movups 48(%rsp), %xmm12 + movups 32(%rsp), %xmm13 + movups 16(%rsp), %xmm14 + movups (%rsp), %xmm15 + movq 136(%rsp), %rsi + movq 128(%rsp), %rdi + movq 168(%rsp), %r12 + cfi_restore (%r12) + movq 160(%rsp), %r13 + cfi_restore (%r13) + movq 152(%rsp), %r14 + cfi_restore (%r14) + movq 144(%rsp), %r15 + cfi_restore (%r15) + movups 256(%rsp), %xmm0 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + movsd 200(%rsp,%r15), %xmm0 + + call cos@PLT + + movsd %xmm0, 264(%rsp,%r15) + jmp .LBL_1_8 + +.LBL_1_12: + movzbl %r12b, %r15d + shlq $4, %r15 + movsd 192(%rsp,%r15), %xmm0 + + call cos@PLT + + movsd %xmm0, 256(%rsp,%r15) + jmp .LBL_1_7 + +END (_ZGVbN2v_cos_sse4) diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S new file mode 100644 index 0000000..5babb83 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core.S @@ -0,0 +1,38 @@ +/* Multiple versions of vectorized cos, vector length is 4. + 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 + . */ + +#include +#include + + .text +ENTRY (_ZGVdN4v_cos) + .type _ZGVdN4v_cos, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1f + call __init_cpu_features +1: leaq _ZGVdN4v_cos_avx2(%rip), %rax + testl $bit_AVX2_Usable, __cpu_features+FEATURE_OFFSET+index_AVX2_Usable(%rip) + jz 2f + ret +2: leaq _ZGVdN4v_cos_sse_wrapper(%rip), %rax + ret +END (_ZGVdN4v_cos) +libmvec_hidden_def (_ZGVdN4v_cos) + +#define _ZGVdN4v_cos _ZGVdN4v_cos_sse_wrapper +#include "../svml_d_cos4_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S new file mode 100644 index 0000000..f192ba0 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos4_core_avx2.S @@ -0,0 +1,207 @@ +/* Function cos 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 + . */ + +#include +#include "svml_d_cos_data.h" + + .text +ENTRY (_ZGVdN4v_cos_avx2) + +/* ALGORITHM DESCRIPTION: + + ( low accuracy ( < 4ulp ) or enhanced performance + ( half of correct mantissa ) implementation ) + + Argument representation: + arg + Pi/2 = (N*Pi + R) + + Result calculation: + cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R) + sin(R) is approximated by corresponding polynomial + */ + 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_dcos_data@GOTPCREL(%rip), %rax + vmovapd %ymm0, %ymm1 + vmovupd __dInvPI(%rax), %ymm4 + vmovupd __dRShifter(%rax), %ymm5 + +/* + ARGUMENT RANGE REDUCTION: + Add Pi/2 to argument: X' = X+Pi/2 + */ + vaddpd __dHalfPI(%rax), %ymm1, %ymm7 + +/* Get absolute argument value: X' = |X'| */ + vandpd __dAbsMask(%rax), %ymm7, %ymm2 + +/* Y = X'*InvPi + RS : right shifter add */ + vfmadd213pd %ymm5, %ymm4, %ymm7 + vmovupd __dC7(%rax), %ymm4 + +/* Check for large arguments path */ + vcmpnle_uqpd __dRangeVal(%rax), %ymm2, %ymm3 + +/* N = Y - RS : right shifter sub */ + vsubpd %ymm5, %ymm7, %ymm6 + vmovupd __dPI1_FMA(%rax), %ymm2 + +/* SignRes = Y<<63 : shift LSB to MSB place for result sign */ + vpsllq $63, %ymm7, %ymm7 + +/* N = N - 0.5 */ + vsubpd __dOneHalf(%rax), %ymm6, %ymm0 + vmovmskpd %ymm3, %ecx + +/* R = X - N*Pi1 */ + vmovapd %ymm1, %ymm3 + vfnmadd231pd %ymm0, %ymm2, %ymm3 + +/* R = R - N*Pi2 */ + vfnmadd231pd __dPI2_FMA(%rax), %ymm0, %ymm3 + +/* R = R - N*Pi3 */ + vfnmadd132pd __dPI3_FMA(%rax), %ymm3, %ymm0 + +/* POLYNOMIAL APPROXIMATION: R2 = R*R */ + vmulpd %ymm0, %ymm0, %ymm5 + vfmadd213pd __dC6(%rax), %ymm5, %ymm4 + vfmadd213pd __dC5(%rax), %ymm5, %ymm4 + vfmadd213pd __dC4(%rax), %ymm5, %ymm4 + +/* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */ + vfmadd213pd __dC3(%rax), %ymm5, %ymm4 + +/* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */ + vfmadd213pd __dC2(%rax), %ymm5, %ymm4 + vfmadd213pd __dC1(%rax), %ymm5, %ymm4 + vmulpd %ymm5, %ymm4, %ymm6 + vfmadd213pd %ymm0, %ymm0, %ymm6 + +/* + RECONSTRUCTION: + Final sign setting: Res = Poly^SignRes */ + vxorpd %ymm7, %ymm6, %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 %ymm1, 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 cos@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 cos@PLT + + vmovsd %xmm0, 384(%rsp,%r15) + jmp .LBL_1_7 + +END (_ZGVdN4v_cos_avx2) diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S new file mode 100644 index 0000000..ba3b66f --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core.S @@ -0,0 +1,39 @@ +/* Multiple versions of vectorized cos, vector length is 8. + 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 + . */ + +#include +#include + + .text +ENTRY (_ZGVeN8v_cos) + .type _ZGVeN8v_cos, @gnu_indirect_function + cmpl $0, KIND_OFFSET+__cpu_features(%rip) + jne 1 + call __init_cpu_features +1: leaq _ZGVeN8v_cos_skx(%rip), %rax + testl $bit_AVX512DQ_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512DQ_Usable(%rip) + jnz 3 +2: leaq _ZGVeN8v_cos_knl(%rip), %rax + testl $bit_AVX512F_Usable, __cpu_features+FEATURE_OFFSET+index_AVX512F_Usable(%rip) + jnz 3 + leaq _ZGVeN8v_cos_avx2_wrapper(%rip), %rax +3: ret +END (_ZGVeN8v_cos) + +#define _ZGVeN8v_cos _ZGVeN8v_cos_avx2_wrapper +#include "../svml_d_cos8_core.S" diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S new file mode 100644 index 0000000..14695ec --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S @@ -0,0 +1,463 @@ +/* Function cos 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 + . */ + +#include +#include "svml_d_cos_data.h" +#include "svml_d_wrapper_impl.h" + + .text +ENTRY (_ZGVeN8v_cos_knl) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN4v_cos +#else +/* + ALGORITHM DESCRIPTION: + + ( low accuracy ( < 4ulp ) or enhanced performance + ( half of correct mantissa ) implementation ) + + Argument representation: + arg + Pi/2 = (N*Pi + R) + + Result calculation: + cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R) + sin(R) is approximated by corresponding polynomial + */ + 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_dcos_data@GOTPCREL(%rip), %rax + +/* R = X - N*Pi1 */ + vmovaps %zmm0, %zmm7 + +/* Check for large arguments path */ + movq $-1, %rcx + +/* + ARGUMENT RANGE REDUCTION: + Add Pi/2 to argument: X' = X+Pi/2 + */ + vaddpd __dHalfPI(%rax), %zmm0, %zmm5 + vmovups __dInvPI(%rax), %zmm3 + +/* Get absolute argument value: X' = |X'| */ + vpandq __dAbsMask(%rax), %zmm5, %zmm1 + +/* Y = X'*InvPi + RS : right shifter add */ + vfmadd213pd __dRShifter(%rax), %zmm3, %zmm5 + vmovups __dPI1_FMA(%rax), %zmm6 + +/* N = Y - RS : right shifter sub */ + vsubpd __dRShifter(%rax), %zmm5, %zmm4 + +/* SignRes = Y<<63 : shift LSB to MSB place for result sign */ + vpsllq $63, %zmm5, %zmm12 + vmovups __dC7(%rax), %zmm8 + +/* N = N - 0.5 */ + vsubpd __dOneHalf(%rax), %zmm4, %zmm10 + vcmppd $22, __dRangeVal(%rax), %zmm1, %k1 + vpbroadcastq %rcx, %zmm2{%k1}{z} + vfnmadd231pd %zmm10, %zmm6, %zmm7 + vptestmq %zmm2, %zmm2, %k0 + +/* R = R - N*Pi2 */ + vfnmadd231pd __dPI2_FMA(%rax), %zmm10, %zmm7 + kmovw %k0, %ecx + movzbl %cl, %ecx + +/* R = R - N*Pi3 */ + vfnmadd132pd __dPI3_FMA(%rax), %zmm7, %zmm10 + +/* + POLYNOMIAL APPROXIMATION: + R2 = R*R + */ + vmulpd %zmm10, %zmm10, %zmm9 + vfmadd213pd __dC6(%rax), %zmm9, %zmm8 + vfmadd213pd __dC5(%rax), %zmm9, %zmm8 + vfmadd213pd __dC4(%rax), %zmm9, %zmm8 + +/* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */ + vfmadd213pd __dC3(%rax), %zmm9, %zmm8 + +/* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */ + vfmadd213pd __dC2(%rax), %zmm9, %zmm8 + vfmadd213pd __dC1(%rax), %zmm9, %zmm8 + vmulpd %zmm9, %zmm8, %zmm11 + vfmadd213pd %zmm10, %zmm10, %zmm11 + +/* + RECONSTRUCTION: + Final sign setting: Res = Poly^SignRes + */ + vpxorq %zmm12, %zmm11, %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 %zmm1, 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), %zmm1 + jmp .LBL_1_2 + +.LBL_1_10: + cfi_restore_state + movzbl %r12b, %r15d + shlq $4, %r15 + vmovsd 1160(%rsp,%r15), %xmm0 + call cos@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 cos@PLT + vmovsd %xmm0, 1216(%rsp,%r15) + jmp .LBL_1_7 +#endif +END (_ZGVeN8v_cos_knl) + +ENTRY (_ZGVeN8v_cos_skx) +#ifndef HAVE_AVX512_ASM_SUPPORT +WRAPPER_IMPL_AVX512 _ZGVdN4v_cos +#else +/* + ALGORITHM DESCRIPTION: + + ( low accuracy ( < 4ulp ) or enhanced performance + ( half of correct mantissa ) implementation ) + + Argument representation: + arg + Pi/2 = (N*Pi + R) + + Result calculation: + cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R) + sin(R) is approximated by corresponding polynomial + */ + 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_dcos_data@GOTPCREL(%rip), %rax + +/* R = X - N*Pi1 */ + vmovaps %zmm0, %zmm8 + +/* Check for large arguments path */ + vpbroadcastq .L_2il0floatpacket.16(%rip), %zmm2 + +/* + ARGUMENT RANGE REDUCTION: + Add Pi/2 to argument: X' = X+Pi/2 + */ + vaddpd __dHalfPI(%rax), %zmm0, %zmm6 + vmovups __dInvPI(%rax), %zmm3 + vmovups __dRShifter(%rax), %zmm4 + vmovups __dPI1_FMA(%rax), %zmm7 + vmovups __dC7(%rax), %zmm9 + +/* Get absolute argument value: X' = |X'| */ + vandpd __dAbsMask(%rax), %zmm6, %zmm1 + +/* Y = X'*InvPi + RS : right shifter add */ + vfmadd213pd %zmm4, %zmm3, %zmm6 + vcmppd $18, __dRangeVal(%rax), %zmm1, %k1 + +/* SignRes = Y<<63 : shift LSB to MSB place for result sign */ + vpsllq $63, %zmm6, %zmm13 + +/* N = Y - RS : right shifter sub */ + vsubpd %zmm4, %zmm6, %zmm5 + +/* N = N - 0.5 */ + vsubpd __dOneHalf(%rax), %zmm5, %zmm11 + vfnmadd231pd %zmm11, %zmm7, %zmm8 + +/* R = R - N*Pi2 */ + vfnmadd231pd __dPI2_FMA(%rax), %zmm11, %zmm8 + +/* R = R - N*Pi3 */ + vfnmadd132pd __dPI3_FMA(%rax), %zmm8, %zmm11 + +/* + POLYNOMIAL APPROXIMATION: + R2 = R*R + */ + vmulpd %zmm11, %zmm11, %zmm10 + vfmadd213pd __dC6(%rax), %zmm10, %zmm9 + vfmadd213pd __dC5(%rax), %zmm10, %zmm9 + vfmadd213pd __dC4(%rax), %zmm10, %zmm9 + +/* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */ + vfmadd213pd __dC3(%rax), %zmm10, %zmm9 + +/* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */ + vfmadd213pd __dC2(%rax), %zmm10, %zmm9 + vfmadd213pd __dC1(%rax), %zmm10, %zmm9 + vmulpd %zmm10, %zmm9, %zmm12 + vfmadd213pd %zmm11, %zmm11, %zmm12 + vpandnq %zmm1, %zmm1, %zmm2{%k1} + vcmppd $3, %zmm2, %zmm2, %k0 + +/* + RECONSTRUCTION: + Final sign setting: Res = Poly^SignRes + */ + vxorpd %zmm13, %zmm12, %zmm1 + kmovw %k0, %ecx + testl %ecx, %ecx + jne .LBL_2_3 + +.LBL_2_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_2_3: + cfi_restore_state + vmovups %zmm0, 1152(%rsp) + vmovups %zmm1, 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), %zmm1 + 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 cos@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 cos@PLT + + vmovsd %xmm0, 1216(%rsp,%r15) + jmp .LBL_2_7 +#endif +END (_ZGVeN8v_cos_skx) + + .section .rodata, "a" +.L_2il0floatpacket.16: + .long 0xffffffff,0xffffffff + .type .L_2il0floatpacket.16,@object -- cgit v1.1