[v1,17/27] x86/fpu: Optimize svml_s_tanf8_core_avx2.S

Message ID	20221207085236.1424424-17-goldstein.w.n@gmail.com
State	New
Headers	show Return-Path: <libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org> DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 4CB99392B132 To: libc-alpha@sourceware.org Cc: goldstein.w.n@gmail.com, hjl.tools@gmail.com, andrey.kolesov@intel.com, carlos@systemhalted.org Subject: [PATCH v1 17/27] x86/fpu: Optimize svml_s_tanf8_core_avx2.S Date: Wed, 7 Dec 2022 00:52:26 -0800 Message-Id: <20221207085236.1424424-17-goldstein.w.n@gmail.com> In-Reply-To: <20221207085236.1424424-1-goldstein.w.n@gmail.com> References: <20221207085236.1424424-1-goldstein.w.n@gmail.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: list From: Noah Goldstein via Libc-alpha <libc-alpha@sourceware.org> Reply-To: Noah Goldstein <goldstein.w.n@gmail.com> Errors-To: libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org Sender: "Libc-alpha" <libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org>
Series	[v1,01/27] x86/fpu: Create helper file for common data macros \| expand [v1,01/27] x86/fpu: Create helper file for common data macros [v1,02/27] x86/fpu: Add file for common data used across svml_s__avx2.S files [v1,03/27] x86/fpu: Add file for common data used across svml_s__avx512.S files [v1,04/27] x86/fpu: Add file for common data used across svml_s__sse4.S files [v1,05/27] x86/fpu: Build common data files for svml_s__{avx512, avx2, sse4}.S [v1,06/27] x86/fpu: Update rodata usage in svml_s_tanhf__{avx2, sse4} [v1,07/27] x86/fpu: Update rodata usage in svml_s_tanhf16_core_avx512.S [v1,08/27] x86/fpu: Update rodata usage in svml_s_atanhf16_core_avx512.S [v1,09/27] x86/fpu: Update rodata usage in svml_s_atanhf4_core_sse4.S [v1,10/27] x86/fpu: Update rodata usage in svml_s_atanhf8_core_avx2.S [v1,11/27] x86/fpu: Optimize svml_s_atanf16_core_avx512.S [v1,12/27] x86/fpu: Optimize svml_s_atanf4_core_sse4.S [v1,13/27] x86/fpu: Optimize svml_s_atanf8_core_avx2.S [v1,14/27] x86/fpu: Add common rodata file for svml_s_tanf__{avx512, avx2, sse4}.S [v1,15/27] x86/fpu: Optimize svml_s_tanf16_core_avx512.S [v1,16/27] x86/fpu: Optimize svml_s_tanf4_core_sse4.S [v1,17/27] x86/fpu: Optimize svml_s_tanf8_core_avx2.S [v1,18/27] x86/fpu: Optimize svml_s_log10f16_core_avx512.S [v1,19/27] x86/fpu: Optimize svml_s_log10f4_core_sse4.S [v1,20/27] x86/fpu: Optimize svml_s_log10f8_core_avx2.S [v1,21/27] x86/fpu: Optimize svml_s_log2f16_core_avx512.S [v1,22/27] x86/fpu: Optimize svml_s_log2f4_core_sse4.S [v1,23/27] x86/fpu: Optimize svml_s_log2f8_core_avx2.S [v1,24/27] x86/fpu: Optimize svml_s_logf16_core_avx512.S [v1,25/27] x86/fpu: Optimize svml_s_logf4_core_sse4.S [v1,26/27] x86/fpu: Optimize svml_s_logf8_core_avx2.S [v1,27/27] x86/fpu: Remove unused svml_s_logf_data.S file

diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_tanf8_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_s_tanf8_core_avx2.S index d34e61ac41..de4c849c45 100644 --- a/sysdeps/x86_64/fpu/multiarch/svml_s_tanf8_core_avx2.S +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_tanf8_core_avx2.S @@ -45,2548 +45,587 @@ * */ -/* Offsets for data table __svml_stan_data_internal - */ -#define _sInvPI_uisa 0 -#define _sPI1_uisa 32 -#define _sPI2_uisa 64 -#define _sPI3_uisa 96 -#define _sPI2_ha_uisa 128 -#define _sPI3_ha_uisa 160 -#define Th_tbl_uisa 192 -#define Tl_tbl_uisa 320 -#define _sPC3_uisa 448 -#define _sPC5_uisa 480 -#define _sRangeReductionVal_uisa 512 -#define _sInvPi 544 -#define _sSignMask 576 -#define _sAbsMask 608 -#define _sRangeVal 640 -#define _sRShifter 672 -#define _sOne 704 -#define _sRangeReductionVal 736 -#define _sPI1 768 -#define _sPI2 800 -#define _sPI3 832 -#define _sPI4 864 -#define _sPI1_FMA 896 -#define _sPI2_FMA 928 -#define _sPI3_FMA 960 -#define _sP0 992 -#define _sP1 1024 -#define _sQ0 1056 -#define _sQ1 1088 -#define _sQ2 1120 -#define _sTwo 1152 -#define _sCoeffs 1184 +#define LOCAL_DATA_NAME __svml_stan_data_internal +#include "svml_s_common_avx2_rodata_offsets.h" + +#define AVX2_SHARED_TABLE +#define AVX512_SHARED_OFFSETS +#include "svml_s_tanf_rodata.h.S" + +/* Offsets for data table __svml_stan_data_internal. */ +#define _sPI2_FMA 0 +#define _sPI3_FMA 32 +#define _FLT_0 64 +#define _FLT_1 96 +#define _FLT_2 128 +#define _FLT_3 160 #include <sysdep.h> .section .text.avx2, "ax", @progbits ENTRY(_ZGVdN8v_tanf_avx2) - pushq %rbp - cfi_def_cfa_offset(16) - movq %rsp, %rbp - cfi_def_cfa(6, 16) - cfi_offset(6, -16) - andq $-32, %rsp - pushq %rbx - subq $184, %rsp - - /* - * Legacy Code - * Here HW FMA can be unavailable - */ - xorl %eax, %eax - vmovups _sAbsMask+__svml_stan_data_internal(%rip), %ymm10 - - /* - * - * Main path (_LA_ and _EP_) - * - * Octant calculation - */ - vmovups _sInvPi+__svml_stan_data_internal(%rip), %ymm5 - vmovups _sRShifter+__svml_stan_data_internal(%rip), %ymm2 - - /* Range reduction */ - vmovups _sPI1_FMA+__svml_stan_data_internal(%rip), %ymm3 - - /* Rational approximation */ - vmovups _sP1+__svml_stan_data_internal(%rip), %ymm9 - vmovaps %ymm0, %ymm12 - vandps %ymm10, %ymm12, %ymm1 + vmovups COMMON_DATA(_AbsMask)(%rip), %ymm6 + /* Main path (_LA_ and _EP_) + Octant calculation. */ + vmovups AVX2_SHARED_DATA(_sInvPi)(%rip), %ymm5 + vmovups AVX2_SHARED_DATA(_sRShifter)(%rip), %ymm2 + + + vandps %ymm6, %ymm0, %ymm1 + vfmadd213ps %ymm2, %ymm1, %ymm5 - vsubps %ymm2, %ymm5, %ymm8 - vpslld $30, %ymm5, %ymm6 - - /* Inversion mask and sign calculation */ - vpslld $31, %ymm5, %ymm4 - vfnmadd213ps %ymm1, %ymm8, %ymm3 - vfnmadd231ps _sPI2_FMA+__svml_stan_data_internal(%rip), %ymm8, %ymm3 - vfnmadd132ps _sPI3_FMA+__svml_stan_data_internal(%rip), %ymm3, %ymm8 - vmovups _sQ2+__svml_stan_data_internal(%rip), %ymm3 - vmulps %ymm8, %ymm8, %ymm13 - vfmadd213ps _sQ1+__svml_stan_data_internal(%rip), %ymm13, %ymm3 - vfmadd213ps _sP0+__svml_stan_data_internal(%rip), %ymm13, %ymm9 - vfmadd213ps _sQ0+__svml_stan_data_internal(%rip), %ymm13, %ymm3 - vmulps %ymm9, %ymm8, %ymm8 - vxorps %ymm7, %ymm7, %ymm7 - vcmpneqps %ymm7, %ymm6, %ymm2 - vandnps %ymm12, %ymm10, %ymm11 - vxorps %ymm11, %ymm4, %ymm0 - - /* Exchanged numerator and denominator if necessary */ - vandnps %ymm8, %ymm2, %ymm14 - vandps %ymm3, %ymm2, %ymm15 - vandps %ymm8, %ymm2, %ymm4 - vandnps %ymm3, %ymm2, %ymm5 - vorps %ymm15, %ymm14, %ymm6 - vorps %ymm5, %ymm4, %ymm7 - - /* Division */ - vdivps %ymm7, %ymm6, %ymm9 - - /* Large values check */ - vcmpnle_uqps _sRangeReductionVal+__svml_stan_data_internal(%rip), %ymm1, %ymm10 - vmovmskps %ymm10, %edx - - /* Sign setting */ - vxorps %ymm0, %ymm9, %ymm0 - - /* - * - * End of main path (_LA_ and _EP_) - */ + vsubps %ymm2, %ymm5, %ymm7 + + /* Range reduction. */ + vmovups COMMON_DATA(_TanSPI1_FMA)(%rip), %ymm3 + vfnmadd213ps %ymm1, %ymm7, %ymm3 + + vfnmadd231ps LOCAL_DATA(_sPI2_FMA)(%rip), %ymm7, %ymm3 + vfnmadd132ps LOCAL_DATA(_sPI3_FMA)(%rip), %ymm3, %ymm7 + vmovups AVX2_SHARED_DATA(_sQ2)(%rip), %ymm3 + /* Rational approximation. */ + vmovups AVX2_SHARED_DATA(_sP1)(%rip), %ymm4 + vmulps %ymm7, %ymm7, %ymm2 + vfmadd213ps AVX2_SHARED_DATA(_sQ1)(%rip), %ymm2, %ymm3 + vmovups AVX2_SHARED_DATA(_sP0)(%rip), %ymm8 + vfmadd213ps %ymm8, %ymm2, %ymm4 + vfmadd213ps %ymm8, %ymm2, %ymm3 + vmulps %ymm4, %ymm7, %ymm4 + /* Inversion mask and sign calculation. */ + vpslld $31, %ymm5, %ymm2 + + + vandnps %ymm0, %ymm6, %ymm7 + + /* Exchanged numerator and denominator if necessary. */ + vblendvps %ymm2, %ymm3, %ymm4, %ymm6 + vblendvps %ymm2, %ymm4, %ymm3, %ymm3 + + /* Large values check. */ + vpcmpgtd AVX2_SHARED_DATA(_sRangeReductionVal)(%rip), %ymm1, %ymm10 + vpmovmskb %ymm10, %edx + + /* Division. */ + vdivps %ymm3, %ymm6, %ymm3 + + /* End of main path (_LA_ and _EP_). */ testl %edx, %edx - /* Go to auxilary branch */ + /* Go to auxilary branch. */ jne L(AUX_BRANCH) - /* DW_CFA_expression: r3 (rbx) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -8; DW_OP_plus) */ - .cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22 - # LOE r12 r13 r14 r15 eax ymm0 ymm1 ymm10 ymm11 ymm12 - /* Return from auxilary branch - * for out of main path inputs - */ + vxorps %ymm2, %ymm7, %ymm7 + /* Sign setting. */ + vxorps %ymm7, %ymm3, %ymm0 + ret -L(AUX_BRANCH_RETURN): - testl %eax, %eax +L(AUX_BRANCH): + /* Sign setting. NB for all special case values this is + equivilent to the input (ymm0). */ + vpandn %ymm3, %ymm10, %ymm3 + vpxor %ymm3, %ymm7, %ymm12 + vmovaps %ymm0, %ymm11 - /* Go to special inputs processing branch */ - jne L(SPECIAL_VALUES_BRANCH) - # LOE r12 r13 r14 r15 eax ymm0 ymm12 - /* Restore registers - * and exit the function - */ + /* Get the (2^a / 2pi) mod 1 values from the table. */ + lea AVX512_SHARED_DATA(_Reduction)(%rip), %rdx -L(EXIT): - addq $184, %rsp - cfi_restore(3) - popq %rbx - movq %rbp, %rsp - popq %rbp - cfi_def_cfa(7, 8) - cfi_restore(6) - ret - cfi_def_cfa(6, 16) - /* DW_CFA_expression: r3 (rbx) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -8; DW_OP_plus) */ - .cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22 - cfi_offset(6, -16) + vpsrld $23, %ymm1, %ymm6 + vpaddd %ymm6, %ymm6, %ymm2 - /* Branch to process - * special inputs - */ + vpaddd %ymm6, %ymm2, %ymm3 -L(SPECIAL_VALUES_BRANCH): - vmovups %ymm12, 32(%rsp) - vmovups %ymm0, 64(%rsp) - # LOE r12 r13 r14 r15 eax ymm0 + /* Collect indexes. */ + vmovq %xmm3, %rax + movl %eax, %ecx + shrq $32, %rax - xorl %ebx, %ebx - # LOE r12 r13 r14 r15 eax ebx + vmovq (%rdx, %rcx, 4), %xmm4 + vmovq (%rdx, %rax, 4), %xmm5 + vpunpckldq %xmm5, %xmm4, %xmm4 - vzeroupper - movq %r12, 8(%rsp) - /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -184; DW_OP_plus) */ - .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 0xff, 0xff, 0xff, 0x22 - movl %eax, %r12d - movq %r13, (%rsp) - /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -192; DW_OP_plus) */ - .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 - # LOE r14 r15 ebx r12d - - /* Range mask - * bits check - */ - -L(RANGEMASK_CHECK): - btl %ebx, %r12d - - /* Call scalar math function */ - jc L(SCALAR_MATH_CALL) - # LOE r14 r15 ebx r12d - - /* Special inputs - * processing loop - */ + vpextrq $1, %xmm3, %rdi + movl %edi, %esi + shrq $32, %rdi -L(SPECIAL_VALUES_LOOP): - incl %ebx - cmpl $8, %ebx - - /* Check bits in range mask */ - jl L(RANGEMASK_CHECK) - # LOE r14 r15 ebx r12d - - movq 8(%rsp), %r12 - cfi_restore(12) - movq (%rsp), %r13 - cfi_restore(13) - vmovups 64(%rsp), %ymm0 - - /* Go to exit */ - jmp L(EXIT) - /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -184; DW_OP_plus) */ - .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 0xff, 0xff, 0xff, 0x22 - /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -192; DW_OP_plus) */ - .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 - # LOE r12 r13 r14 r15 ymm0 - - /* Scalar math fucntion call - * to process special input - */ - -L(SCALAR_MATH_CALL): - movl %ebx, %r13d - vmovss 32(%rsp, %r13, 4), %xmm0 - call tanf@PLT - # LOE r13 r14 r15 ebx r12d xmm0 + vmovq (%rdx, %rsi, 4), %xmm2 + vmovq (%rdx, %rdi, 4), %xmm5 + vpunpckldq %xmm5, %xmm2, %xmm2 - vmovss %xmm0, 64(%rsp, %r13, 4) + vextractf128 $1, %ymm3, %xmm7 - /* Process special inputs in loop */ - jmp L(SPECIAL_VALUES_LOOP) - cfi_restore(12) - cfi_restore(13) - # LOE r14 r15 ebx r12d + vmovq %xmm7, %r10 + movl %r10d, %r8d + shrq $32, %r10 - /* Auxilary branch - * for out of main path inputs - */ + vmovq (%rdx, %r8, 4), %xmm3 + vmovq (%rdx, %r10, 4), %xmm5 + vpunpckldq %xmm5, %xmm3, %xmm3 -L(AUX_BRANCH): - vpand .FLT_16(%rip), %ymm1, %ymm5 - - /* - * Get the (2^a / 2pi) mod 1 values from the table. - * Because doesn't have I-type gather, we need a trivial cast - */ - lea __svml_stan_reduction_data_internal(%rip), %rdx - vmovups %ymm11, 64(%rsp) - vmovups .FLT_15(%rip), %ymm7 - vmovups %ymm10, 96(%rsp) - vmovups %ymm0, 128(%rsp) - vpsrld $23, %ymm5, %ymm6 - vpslld $1, %ymm6, %ymm11 - vpaddd %ymm6, %ymm11, %ymm13 - vpslld $2, %ymm13, %ymm15 - vandps %ymm7, %ymm12, %ymm14 - vcmpeqps %ymm7, %ymm14, %ymm10 - vmovmskps %ymm10, %eax - vextractf128 $1, %ymm15, %xmm7 - vmovd %xmm15, %ecx - vmovd %xmm7, %r8d - vmovd (%rcx, %rdx), %xmm8 - vpextrd $1, %xmm15, %ebx - vpextrd $2, %xmm15, %esi - vpextrd $3, %xmm15, %edi - vpextrd $1, %xmm7, %r10d - vpextrd $2, %xmm7, %r9d - vpextrd $3, %xmm7, %r11d - vmovd (%rbx, %rdx), %xmm3 - vmovd (%rsi, %rdx), %xmm2 - vmovd (%rdi, %rdx), %xmm14 - vmovd (%r8, %rdx), %xmm10 - vmovd (%r10, %rdx), %xmm5 - vmovd (%r9, %rdx), %xmm11 - vmovd (%r11, %rdx), %xmm6 - vpunpckldq %xmm3, %xmm8, %xmm4 - vpunpckldq %xmm14, %xmm2, %xmm0 - vpunpckldq %xmm5, %xmm10, %xmm13 - vpunpckldq %xmm6, %xmm11, %xmm15 - vpunpcklqdq %xmm0, %xmm4, %xmm9 - vmovd 4(%rcx, %rdx), %xmm3 - vmovd 4(%rbx, %rdx), %xmm2 - vmovd 4(%rsi, %rdx), %xmm14 - vmovd 4(%rdi, %rdx), %xmm4 - vpunpcklqdq %xmm15, %xmm13, %xmm8 - vmovd 4(%r8, %rdx), %xmm5 - vmovd 4(%r10, %rdx), %xmm6 - vmovd 4(%r9, %rdx), %xmm13 - vmovd 4(%r11, %rdx), %xmm15 - vpunpckldq %xmm2, %xmm3, %xmm0 - vpunpckldq %xmm4, %xmm14, %xmm7 - vpunpckldq %xmm15, %xmm13, %xmm3 - vpunpcklqdq %xmm7, %xmm0, %xmm10 - vmovd 8(%rsi, %rdx), %xmm0 - vmovd 8(%rdi, %rdx), %xmm7 - vmovd 8(%rcx, %rdx), %xmm14 - vmovd 8(%rbx, %rdx), %xmm4 - vmovd 8(%r8, %rdx), %xmm15 - vinsertf128 $1, %xmm8, %ymm9, %ymm11 - vpunpckldq %xmm6, %xmm5, %xmm8 - vpunpcklqdq %xmm3, %xmm8, %xmm2 - vpunpckldq %xmm7, %xmm0, %xmm6 - - /* - * Also get the significand as an integer - * NB: adding in the integer bit is wrong for denorms! - * To make this work for denorms we should do something slightly different - */ - vpand .FLT_17(%rip), %ymm1, %ymm7 - vmovd 8(%r10, %rdx), %xmm8 - vmovd 8(%r9, %rdx), %xmm3 - vpunpckldq %xmm4, %xmm14, %xmm5 - vpunpckldq %xmm8, %xmm15, %xmm14 - - /* Load constants (not all needed at once) */ - lea _sCoeffs+36+__svml_stan_data_internal(%rip), %r9 - vpunpcklqdq %xmm6, %xmm5, %xmm13 - vpaddd .FLT_18(%rip), %ymm7, %ymm5 - vinsertf128 $1, %xmm2, %ymm10, %ymm9 - vmovd 8(%r11, %rdx), %xmm2 - vpunpckldq %xmm2, %xmm3, %xmm4 - vpunpcklqdq %xmm4, %xmm14, %xmm0 - - /* - * Break the P_xxx and m into 16-bit chunks ready for - * the long multiplication via 16x16->32 multiplications - */ - vmovdqu .FLT_19(%rip), %ymm14 - vpsrld $16, %ymm5, %ymm10 - vpand %ymm14, %ymm5, %ymm5 - vpand %ymm14, %ymm9, %ymm3 - vpand %ymm14, %ymm11, %ymm7 - vpsrld $16, %ymm11, %ymm11 - vpmulld %ymm3, %ymm5, %ymm8 - vpmulld %ymm3, %ymm10, %ymm3 + vpextrq $1, %xmm7, %r11 + movl %r11d, %r9d + shrq $32, %r11 + + vmovq (%rdx, %r9, 4), %xmm7 + vmovq (%rdx, %r11, 4), %xmm5 + vpunpckldq %xmm5, %xmm7, %xmm7 + + vinsertf128 $1, %xmm3, %ymm4, %ymm4 + vinsertf128 $1, %xmm7, %ymm2, %ymm2 + + vmovdqa LOCAL_DATA(_FLT_0)(%rip), %ymm9 + + + vpunpcklqdq %ymm2, %ymm4, %ymm7 + vpunpckhqdq %ymm2, %ymm4, %ymm6 + + /* Break the P_xxx and m into 16-bit chunks ready for + the long multiplication via 16x16->32 multiplications. */ + vpandn %ymm1, %ymm9, %ymm5 + vpsrld $16, %ymm5, %ymm3 + + vpor LOCAL_DATA(_FLT_1)(%rip), %ymm3, %ymm4 + vmovd 8(%rdx, %rcx, 4), %xmm5 + vmovd 8(%rdx, %rax, 4), %xmm2 + vpunpckldq %xmm2, %xmm5, %xmm3 + + vmovd 8(%rdx, %rsi, 4), %xmm2 + vmovd 8(%rdx, %rdi, 4), %xmm5 + vpunpckldq %xmm5, %xmm2, %xmm2 + + vpunpcklqdq %xmm2, %xmm3, %xmm13 + + vmovd 8(%rdx, %r8, 4), %xmm3 + vmovd 8(%rdx, %r10, 4), %xmm5 + vpunpckldq %xmm5, %xmm3, %xmm0 + /* Also get the significand as an integer + NB: adding in the integer bit is wrong for denorms! + To make this work for denorms we should do something + slightly different. */ + vmovd 8(%rdx, %r9, 4), %xmm2 + vmovd 8(%rdx, %r11, 4), %xmm5 + vpunpckldq %xmm5, %xmm2, %xmm2 + + /* Better to use `vpand` than `vpblendw`. */ + vmovdqu AVX2_SHARED_DATA(_Low16)(%rip), %ymm3 + + vpunpcklqdq %xmm2, %xmm0, %xmm0 vinsertf128 $1, %xmm0, %ymm13, %ymm13 + + vpand %ymm3, %ymm1, %ymm5 + vpand %ymm3, %ymm6, %ymm0 + vpsrld $16, %ymm7, %ymm2 + vpand %ymm3, %ymm7, %ymm7 + vpmulld %ymm0, %ymm5, %ymm8 + vpmulld %ymm0, %ymm4, %ymm14 + vpsrld $16, %ymm6, %ymm0 + vpmulld %ymm2, %ymm5, %ymm2 + vpand %ymm3, %ymm2, %ymm15 vpsrld $16, %ymm13, %ymm6 - vpand %ymm14, %ymm13, %ymm15 - vpsrld $16, %ymm9, %ymm0 - vpmulld %ymm6, %ymm10, %ymm13 + vpand %ymm3, %ymm13, %ymm2 + vpmulld %ymm6, %ymm4, %ymm13 vpmulld %ymm6, %ymm5, %ymm6 vpsrld $16, %ymm6, %ymm6 - vpmulld %ymm15, %ymm10, %ymm4 - vpand %ymm14, %ymm8, %ymm15 + vpmulld %ymm2, %ymm4, %ymm2 vpaddd %ymm6, %ymm13, %ymm13 - vpsrld $16, %ymm4, %ymm4 + vpsrld $16, %ymm2, %ymm6 + vpand %ymm3, %ymm8, %ymm2 vpsrld $16, %ymm8, %ymm8 - vpaddd %ymm13, %ymm15, %ymm15 + vpaddd %ymm13, %ymm2, %ymm13 vpmulld %ymm0, %ymm5, %ymm2 - vpaddd %ymm15, %ymm4, %ymm13 - vpand %ymm14, %ymm2, %ymm4 - vpaddd %ymm8, %ymm3, %ymm15 + vpaddd %ymm13, %ymm6, %ymm13 + vpaddd %ymm8, %ymm14, %ymm14 + vpand %ymm3, %ymm2, %ymm8 vpsrld $16, %ymm2, %ymm2 vpsrld $16, %ymm13, %ymm6 - /* Assemble reduced argument from the pieces */ - vpand %ymm14, %ymm13, %ymm13 - vpaddd %ymm15, %ymm4, %ymm8 - vpmulld %ymm7, %ymm5, %ymm9 - vpmulld %ymm0, %ymm10, %ymm0 - vpaddd %ymm8, %ymm6, %ymm4 - vpand %ymm14, %ymm9, %ymm6 - vpaddd %ymm2, %ymm0, %ymm8 - vpsrld $16, %ymm9, %ymm3 - vpsrld $16, %ymm4, %ymm15 - vpslld $16, %ymm4, %ymm4 - vpaddd %ymm8, %ymm6, %ymm6 - vpaddd %ymm6, %ymm15, %ymm0 - vpmulld %ymm11, %ymm5, %ymm6 - - /* Now do the big multiplication and carry propagation */ - vpmulld %ymm7, %ymm10, %ymm8 - vpand %ymm14, %ymm6, %ymm2 - vpaddd %ymm3, %ymm8, %ymm5 - vpsrld $16, %ymm0, %ymm15 - vpand %ymm14, %ymm0, %ymm0 - - /* - * We want to incorporate the original sign now too. - * Do it here for convenience in getting the right N value, - * though we could wait right to the end if we were prepared - * to modify the sign of N later too. - * So get the appropriate sign mask now (or sooner). - */ - vpand .FLT_20(%rip), %ymm1, %ymm3 - vpaddd %ymm5, %ymm2, %ymm7 - vpaddd %ymm13, %ymm4, %ymm8 - - /* - * Now round at the 2^-8 bit position for reduction mod pi/2^7 - * instead of the original 2pi (but still with the same 2pi scaling). - * Use a shifter of 2^15 + 2^14. - * The N we get is our final version; it has an offset of - * 2^8 because of the implicit integer bit, and anyway for negative - * starting value it's a 2s complement thing. But we need to mask - * off the exponent part anyway so it's fine. - */ - vmovups .FLT_22(%rip), %ymm14 - vpaddd %ymm7, %ymm15, %ymm15 - - /* - * Create floating-point high part, implicitly adding integer bit 1 - * Incorporate overall sign at this stage too. - */ - vpxor .FLT_21(%rip), %ymm3, %ymm11 - - /* - * Create floating-point low and medium parts, respectively - * lo_17, ... lo_0, 0, ..., 0 - * hi_8, ... hi_0, lo_31, ..., lo_18 - * then subtract off the implicitly added integer bits, - * 2^-46 and 2^-23, respectively. - * Put the original sign into all of them at this stage. - */ - vpxor .FLT_23(%rip), %ymm3, %ymm7 - vpslld $16, %ymm15, %ymm9 - vpaddd %ymm0, %ymm9, %ymm2 - vpand .FLT_24(%rip), %ymm8, %ymm0 + /* Assemble reduced argument from the pieces. */ + vpand %ymm3, %ymm13, %ymm13 + vpaddd %ymm14, %ymm8, %ymm8 + vpmulld %ymm7, %ymm5, %ymm5 + vpmulld %ymm0, %ymm4, %ymm0 + vpaddd %ymm8, %ymm6, %ymm8 + vpand %ymm3, %ymm5, %ymm6 + vpaddd %ymm2, %ymm0, %ymm0 + vpsrld $16, %ymm5, %ymm14 + vpsrld $16, %ymm8, %ymm5 + vpslld $16, %ymm8, %ymm8 + vpaddd %ymm0, %ymm6, %ymm6 + vpaddd %ymm6, %ymm5, %ymm0 + + + /* Now do the big multiplication and carry propagation. */ + vpmulld %ymm7, %ymm4, %ymm7 + vpaddd %ymm14, %ymm7, %ymm5 + vpsrld $16, %ymm0, %ymm2 + vpand %ymm3, %ymm0, %ymm0 + + vpaddd %ymm5, %ymm15, %ymm7 + vpaddd %ymm13, %ymm8, %ymm8 + + /* Now round at the 2^-8 bit position for reduction mod pi/2^7 + instead of the original 2pi (but still with the same 2pi scaling). + Use a shifter of 2^15 + 2^14. + The N we get is our final version; it has an offset of + 2^8 because of the implicit integer bit, and anyway for negative + starting value it's a 2s complement thing. But we need to mask + off the exponent part anyway so it's fine. */ + vpaddd %ymm7, %ymm2, %ymm2 + vmovups AVX2_SHARED_DATA(_SH_FLT_1)(%rip), %ymm14 + + /* Create floating-point low and medium parts, respectively + lo_17, ... lo_0, 0, ..., 0 + hi_8, ... hi_0, lo_31, ..., lo_18 + then subtract off the implicitly added integer bits, + 2^-46 and 2^-23, respectively. + Put the original sign into all of them at this stage. */ + vmovdqa AVX2_SHARED_DATA(_SH_FLT_2)(%rip), %ymm7 + + vpslld $16, %ymm2, %ymm2 + vpaddd %ymm0, %ymm2, %ymm2 + vpand AVX2_SHARED_DATA(_Low18)(%rip), %ymm8, %ymm0 vpsrld $18, %ymm8, %ymm8 - vpsrld $9, %ymm2, %ymm10 + vpsrld $9, %ymm2, %ymm6 vpslld $5, %ymm0, %ymm4 - vpor %ymm11, %ymm10, %ymm6 - vpxor .FLT_25(%rip), %ymm3, %ymm11 - vpand .FLT_26(%rip), %ymm2, %ymm3 + vmovdqa COMMON_DATA(_OneF)(%rip), %ymm15 + vpor %ymm15, %ymm6, %ymm6 + + + vpand AVX2_SHARED_DATA(_Low9)(%rip), %ymm2, %ymm3 vpor %ymm7, %ymm4, %ymm5 + vmovdqa AVX2_SHARED_DATA(_SH_FLT_3)(%rip), %ymm4 - /* - * If the magnitude of the input is <= 2^-20, then - * just pass through the input, since no reduction will be needed and - * the main path will only work accurately if the reduced argument is - * about >= 2^-40 (which it is for all large pi multiples) - */ - vmovups .FLT_30(%rip), %ymm4 + /* If the magnitude of the input is <= 2^-20, then + just pass through the input, since no reduction will be needed + and the main path will only work accurately if the reduced + argument is about >= 2^-40 (which it is for all large pi + multiples). */ vpslld $14, %ymm3, %ymm2 - /* - * Now multiply those numbers all by 2 pi, reasonably accurately. - * (RHi + RLo) * (pi_lead + pi_trail) ~= - * RHi * pi_lead + (RHi * pi_trail + RLo * pi_lead) - */ - vmovups .FLT_27(%rip), %ymm3 + /* Now multiply those numbers all by 2 pi, reasonably accurately. + (RHi + RLo) * (pi_lead + pi_trail) ~= + RHi * pi_lead + (RHi * pi_trail + RLo * pi_lead). */ + vmovups AVX2_SHARED_DATA(_SH_FLT_4)(%rip), %ymm3 vaddps %ymm14, %ymm6, %ymm13 - vpor %ymm8, %ymm2, %ymm9 - vsubps %ymm14, %ymm13, %ymm15 - - /* Grab our final N value as an integer, appropriately masked mod 2^8 */ - vpand .FLT_31(%rip), %ymm13, %ymm13 - vpor %ymm11, %ymm9, %ymm10 - vsubps %ymm15, %ymm6, %ymm6 - vsubps %ymm7, %ymm5, %ymm15 - vsubps %ymm11, %ymm10, %ymm14 - - /* Now add them up into 2 reasonably aligned pieces */ + vpor %ymm8, %ymm2, %ymm2 + vsubps %ymm14, %ymm13, %ymm0 + + /* Grab our final N value as an integer, appropriately masked + mod 2^8. */ + vpor %ymm4, %ymm2, %ymm2 + vsubps %ymm0, %ymm6, %ymm6 + vsubps %ymm7, %ymm5, %ymm0 + vsubps %ymm4, %ymm2, %ymm14 + + vmovups LOCAL_DATA(_FLT_2)(%rip), %ymm4 + /* Now add them up into 2 reasonably aligned pieces. */ vaddps %ymm14, %ymm6, %ymm2 vsubps %ymm2, %ymm6, %ymm6 vmulps %ymm2, %ymm3, %ymm7 vaddps %ymm6, %ymm14, %ymm8 - vaddps %ymm8, %ymm15, %ymm8 - vmovaps %ymm3, %ymm15 - vfmsub213ps %ymm7, %ymm2, %ymm15 - vandps .FLT_29(%rip), %ymm1, %ymm0 - vfmadd132ps .FLT_28(%rip), %ymm15, %ymm2 - vcmpgt_oqps %ymm4, %ymm0, %ymm9 - vcmple_oqps %ymm4, %ymm0, %ymm5 - - /* - * The output is _VRES_R (high) + _VRES_E (low), and the integer part is _VRES_IND - * Set sRp2 = _VRES_R^2 and then resume the original code. - * Argument reduction is now finished: x = n * pi/128 + r - * where n = iIndex and r = sR (high) + sE (low). - * But we have n modulo 256, needed for sin/cos with period 2pi - * but we want it modulo 128 since tan has period pi. - */ - vpand .FLT_32(%rip), %ymm13, %ymm0 + vaddps %ymm8, %ymm0, %ymm8 + vmovaps %ymm3, %ymm0 + vfmsub213ps %ymm7, %ymm2, %ymm0 + + + vfmadd132ps LOCAL_DATA(_FLT_3)(%rip), %ymm0, %ymm2 + vpcmpgtd %ymm4, %ymm1, %ymm5 + + /* The output is _VRES_R (high) + _VRES_E (low), and the integer + part is _VRES_IND Set sRp2 = _VRES_R^2 and then resume the + original code. Argument reduction is now finished: x = n * + pi/128 + r where n = iIndex and r = sR (high) + sE (low). + But we have n modulo 256, needed for sin/cos with period 2pi + but we want it modulo 128 since tan has period pi. */ + vpand AVX2_SHARED_DATA(_Low7)(%rip), %ymm13, %ymm0 vfmadd213ps %ymm2, %ymm3, %ymm8 + + + vpslld $2, %ymm0, %ymm2 - vandps %ymm1, %ymm5, %ymm1 - vandps %ymm7, %ymm9, %ymm6 - vorps %ymm6, %ymm1, %ymm15 - vpaddd %ymm0, %ymm2, %ymm1 - vpslld $3, %ymm1, %ymm4 - vandps %ymm8, %ymm9, %ymm3 - - /* - * Simply combine the two parts of the reduced argument - * since we can afford a few ulps in this case. - */ - vaddps %ymm3, %ymm15, %ymm6 - vextractf128 $1, %ymm4, %xmm8 - vmovd %xmm4, %r10d - vmovd %xmm8, %ebx - vmovd -36(%r10, %r9), %xmm5 - vmovd -32(%r10, %r9), %xmm9 - vpextrd $1, %xmm4, %r8d - vpextrd $2, %xmm4, %edi - vpextrd $3, %xmm4, %esi - vpextrd $1, %xmm8, %ecx - vpextrd $2, %xmm8, %edx - vpextrd $3, %xmm8, %r11d - vmovd -36(%r8, %r9), %xmm7 - vmovd -36(%rdi, %r9), %xmm10 - vmovd -36(%rsi, %r9), %xmm11 - vmovd -36(%rbx, %r9), %xmm3 - vmovd -36(%rcx, %r9), %xmm2 - vmovd -36(%rdx, %r9), %xmm0 - vmovd -36(%r11, %r9), %xmm1 - vpunpckldq %xmm7, %xmm5, %xmm14 - vpunpckldq %xmm11, %xmm10, %xmm13 - vpunpckldq %xmm2, %xmm3, %xmm4 - vpunpckldq %xmm1, %xmm0, %xmm5 - vpunpcklqdq %xmm13, %xmm14, %xmm15 - vpunpcklqdq %xmm5, %xmm4, %xmm7 - vmovd -32(%r8, %r9), %xmm10 - vmovd -32(%rdi, %r9), %xmm11 - vmovd -32(%rsi, %r9), %xmm14 - vmovd -32(%rbx, %r9), %xmm2 - vmovd -32(%rcx, %r9), %xmm0 - vmovd -32(%rdx, %r9), %xmm1 - vmovd -32(%r11, %r9), %xmm4 - vpunpckldq %xmm14, %xmm11, %xmm8 - vpunpckldq %xmm0, %xmm2, %xmm5 - vmovd -28(%r8, %r9), %xmm11 - vmovd -28(%rdi, %r9), %xmm14 - vinsertf128 $1, %xmm7, %ymm15, %ymm13 - vpunpckldq %xmm10, %xmm9, %xmm15 - vpunpckldq %xmm4, %xmm1, %xmm7 - vpunpcklqdq %xmm8, %xmm15, %xmm3 - vpunpcklqdq %xmm7, %xmm5, %xmm9 - vmovd -28(%r10, %r9), %xmm10 - vmovd -28(%rsi, %r9), %xmm8 - vmovd -28(%rbx, %r9), %xmm1 - vmovd -28(%rcx, %r9), %xmm4 - vmovd -28(%rdx, %r9), %xmm5 - vmovd -28(%r11, %r9), %xmm7 - vpunpckldq %xmm8, %xmm14, %xmm2 - vmovd -24(%r10, %r9), %xmm14 - vinsertf128 $1, %xmm9, %ymm3, %ymm15 - vpunpckldq %xmm11, %xmm10, %xmm3 - vpunpckldq %xmm4, %xmm1, %xmm9 - vpunpckldq %xmm7, %xmm5, %xmm10 - vpunpcklqdq %xmm2, %xmm3, %xmm0 - vpunpcklqdq %xmm10, %xmm9, %xmm11 - vmovd -24(%r8, %r9), %xmm3 - vmovd -24(%rdi, %r9), %xmm2 - vmovd -24(%rbx, %r9), %xmm7 - vmovd -24(%rcx, %r9), %xmm9 - vmovd -24(%rdx, %r9), %xmm10 - vpunpckldq %xmm3, %xmm14, %xmm1 - vpunpckldq %xmm9, %xmm7, %xmm14 - vmovd -20(%rsi, %r9), %xmm7 - vinsertf128 $1, %xmm11, %ymm0, %ymm8 - vmovd -24(%rsi, %r9), %xmm0 - vmovd -24(%r11, %r9), %xmm11 - vpunpckldq %xmm0, %xmm2, %xmm4 - vpunpckldq %xmm11, %xmm10, %xmm3 - vpunpcklqdq %xmm4, %xmm1, %xmm5 - vpunpcklqdq %xmm3, %xmm14, %xmm2 - vmovd -20(%r10, %r9), %xmm0 - vmovd -20(%r8, %r9), %xmm1 - vmovd -20(%rbx, %r9), %xmm14 - vmovd -20(%rdi, %r9), %xmm4 - vpunpckldq %xmm1, %xmm0, %xmm9 - vmovd -20(%r11, %r9), %xmm0 - vpunpckldq %xmm7, %xmm4, %xmm10 - vpunpcklqdq %xmm10, %xmm9, %xmm11 - vmovd -16(%r10, %r9), %xmm9 - vmovd -16(%r8, %r9), %xmm10 - vinsertf128 $1, %xmm2, %ymm5, %ymm3 - vmovd -20(%rcx, %r9), %xmm2 - vpunpckldq %xmm2, %xmm14, %xmm1 - vmovd -20(%rdx, %r9), %xmm14 - vpunpckldq %xmm0, %xmm14, %xmm4 - vpunpcklqdq %xmm4, %xmm1, %xmm5 - vmovd -16(%rdi, %r9), %xmm2 - vmovd -16(%rsi, %r9), %xmm0 - vpunpckldq %xmm10, %xmm9, %xmm1 - vmovd -16(%rcx, %r9), %xmm9 - vmovd -16(%rdx, %r9), %xmm10 - vpunpckldq %xmm0, %xmm2, %xmm4 - vinsertf128 $1, %xmm5, %ymm11, %ymm7 - vmovups %ymm7, 32(%rsp) - vmovd -16(%rbx, %r9), %xmm7 - vmovd -16(%r11, %r9), %xmm11 - vpunpckldq %xmm9, %xmm7, %xmm14 - vpunpckldq %xmm11, %xmm10, %xmm2 - vpunpcklqdq %xmm4, %xmm1, %xmm5 - vpunpcklqdq %xmm2, %xmm14, %xmm0 - vmovd -12(%r10, %r9), %xmm1 - vmovd -12(%r8, %r9), %xmm4 - vmovd -12(%rdi, %r9), %xmm7 - vmovd -12(%rsi, %r9), %xmm9 - vpunpckldq %xmm4, %xmm1, %xmm10 - vmovd -12(%rcx, %r9), %xmm1 - vmovd -12(%rdx, %r9), %xmm4 - vpunpckldq %xmm9, %xmm7, %xmm11 - vpunpcklqdq %xmm11, %xmm10, %xmm14 - vinsertf128 $1, %xmm0, %ymm5, %ymm2 - vmovd -12(%rbx, %r9), %xmm0 - vmovd -12(%r11, %r9), %xmm5 - vpunpckldq %xmm1, %xmm0, %xmm7 - vpunpckldq %xmm5, %xmm4, %xmm9 - vpunpcklqdq %xmm9, %xmm7, %xmm10 - vmovd -8(%r10, %r9), %xmm1 - vmovd -8(%r8, %r9), %xmm4 - vmovups 128(%rsp), %ymm0 - vinsertf128 $1, %xmm10, %ymm14, %ymm11 - vmovups %ymm11, (%rsp) - vmovups 96(%rsp), %ymm10 - vmovups 64(%rsp), %ymm11 - # LOE rdx rcx rbx rsi rdi r8 r9 r10 r11 r12 r13 r14 r15 eax xmm1 xmm4 ymm0 ymm2 ymm3 ymm6 ymm8 ymm10 ymm11 ymm12 ymm13 ymm15 - - vmovd -8(%rdi, %r9), %xmm7 - vmovd -8(%rsi, %r9), %xmm5 - vpunpckldq %xmm4, %xmm1, %xmm4 - vpunpckldq %xmm5, %xmm7, %xmm9 - vpunpcklqdq %xmm9, %xmm4, %xmm7 - vmovd -8(%rbx, %r9), %xmm1 - vmovd -8(%rcx, %r9), %xmm14 - vmovd -8(%rdx, %r9), %xmm5 - vmovd -8(%r11, %r9), %xmm4 - vpunpckldq %xmm14, %xmm1, %xmm9 - vpunpckldq %xmm4, %xmm5, %xmm1 - vpunpcklqdq %xmm1, %xmm9, %xmm14 - vmovd -4(%r10, %r9), %xmm5 - vmovd -4(%r8, %r9), %xmm4 - vmovd -4(%rdi, %r9), %xmm9 - vmovd -4(%rsi, %r9), %xmm1 - vinsertf128 $1, %xmm14, %ymm7, %ymm7 - vpunpckldq %xmm4, %xmm5, %xmm14 - vpunpckldq %xmm1, %xmm9, %xmm5 - vpunpcklqdq %xmm5, %xmm14, %xmm4 - vmovd -4(%rbx, %r9), %xmm9 - vmovd -4(%rcx, %r9), %xmm1 - vmovd -4(%rdx, %r9), %xmm14 - vmovd -4(%r11, %r9), %xmm5 - vpunpckldq %xmm1, %xmm9, %xmm9 - vpunpckldq %xmm5, %xmm14, %xmm1 - vpunpcklqdq %xmm1, %xmm9, %xmm14 - vmovd (%r10, %r9), %xmm5 - vmovd (%r8, %r9), %xmm9 - vmovd (%rdi, %r9), %xmm1 - vpunpckldq %xmm9, %xmm5, %xmm5 - - /* - * Higher polynomial terms - * Stage 1 (with unlimited parallelism) - * P3 = C1_lo + C2 * Z - */ - vfmadd213ps (%rsp), %ymm6, %ymm7 - vinsertf128 $1, %xmm14, %ymm4, %ymm4 - vmovd (%rsi, %r9), %xmm14 - vpunpckldq %xmm14, %xmm1, %xmm9 - vmovd (%rbx, %r9), %xmm1 - vmovd (%rcx, %r9), %xmm14 - vpunpcklqdq %xmm9, %xmm5, %xmm9 - vpunpckldq %xmm14, %xmm1, %xmm5 - vmovd (%rdx, %r9), %xmm1 - vmovd (%r11, %r9), %xmm14 - vpunpckldq %xmm14, %xmm1, %xmm1 - vpunpcklqdq %xmm1, %xmm5, %xmm5 - vmovups .FLT_33(%rip), %ymm1 - - /* - * Compute 2-part reciprocal component - * Construct a separate reduced argument modulo pi near pi/2 multiples. - * i.e. (pi/2 - x) mod pi, simply by subtracting the reduced argument - * from an accurate B_hi + B_lo = (128 - n) pi/128. Force the upper part - * of this reduced argument to half-length to simplify accurate - * reciprocation later on. - */ - vsubps %ymm6, %ymm13, %ymm14 - vsubps %ymm14, %ymm13, %ymm13 + vpaddd %ymm0, %ymm2, %ymm4 + + vpblendvb %ymm5, %ymm7, %ymm1, %ymm6 + + vandps %ymm8, %ymm5, %ymm3 + vaddps %ymm3, %ymm6, %ymm6 + + + + + /* Simply combine the two parts of the reduced argument + since we can afford a few ulps in this case. */ + + /* Load constants (not all needed at once). */ + lea AVX2_SHARED_DATA(_Coeffs)(%rip), %rdx + + vmovq %xmm4, %rcx + movl %ecx, %eax + shrq $32, %rcx + + vmovdqu (%rdx, %rax, 8), %ymm5 + vmovdqu (%rdx, %rcx, 8), %ymm7 + vpunpckldq %ymm7, %ymm5, %ymm3 + vpunpckhdq %ymm7, %ymm5, %ymm7 + + vpextrq $1, %xmm4, %rsi + movl %esi, %edi + shrq $32, %rsi + + vmovdqu (%rdx, %rdi, 8), %ymm5 + vmovdqu (%rdx, %rsi, 8), %ymm2 + vpunpckldq %ymm2, %ymm5, %ymm0 + vpunpckhdq %ymm2, %ymm5, %ymm2 + + vextractf128 $1, %ymm4, %xmm4 + + vmovq %xmm4, %r8 + movl %r8d, %r10d + shrq $32, %r8 + + vmovdqu (%rdx, %r10, 8), %ymm8 + vmovdqu (%rdx, %r8, 8), %ymm5 + vpunpckldq %ymm5, %ymm8, %ymm14 + vpunpckhdq %ymm5, %ymm8, %ymm8 + + + vpextrq $1, %xmm4, %r11 + movl %r11d, %r9d + shrq $32, %r11 + + vmovdqu (%rdx, %r9, 8), %ymm5 + vmovdqu (%rdx, %r11, 8), %ymm4 + + vpunpckldq %ymm4, %ymm5, %ymm13 + vpunpckhdq %ymm4, %ymm5, %ymm4 + + vpunpcklqdq %ymm0, %ymm3, %ymm5 + vpunpckhqdq %ymm0, %ymm3, %ymm3 + + vpunpcklqdq %ymm13, %ymm14, %ymm0 + vpunpckhqdq %ymm13, %ymm14, %ymm14 + + vinserti128 $0x1, %xmm0, %ymm5, %ymm13 + vperm2i128 $0x31, %ymm0, %ymm5, %ymm5 + + vinserti128 $0x1, %xmm14, %ymm3, %ymm0 + vperm2i128 $0x31, %ymm14, %ymm3, %ymm14 + + vpunpcklqdq %ymm2, %ymm7, %ymm3 + vpunpckhqdq %ymm2, %ymm7, %ymm2 + + vpunpcklqdq %ymm4, %ymm8, %ymm7 + vpunpckhqdq %ymm4, %ymm8, %ymm4 + + vinserti128 $0x1, %xmm7, %ymm3, %ymm8 + vperm2i128 $0x31, %ymm7, %ymm3, %ymm3 + + vperm2i128 $0x31, %ymm4, %ymm2, %ymm7 + vfmadd213ps %ymm3, %ymm6, %ymm7 + vinserti128 $0x1, %xmm4, %ymm2, %ymm3 + + /* Compute 2-part reciprocal component Construct a separate + reduced argument modulo pi near pi/2 multiples. i.e. (pi/2 - + x) mod pi, simply by subtracting the reduced argument from + an accurate B_hi + B_lo = (128 - n) pi/128. Force the upper + part of this reduced argument to half-length to simplify + accurate reciprocation later on. */ + vsubps %ymm6, %ymm13, %ymm2 + vsubps %ymm2, %ymm13, %ymm13 vsubps %ymm6, %ymm13, %ymm13 - vinsertf128 $1, %xmm5, %ymm9, %ymm5 - vandps %ymm1, %ymm14, %ymm9 - vsubps %ymm9, %ymm14, %ymm14 - - /* P4 = C3 + C4 * Z */ - vfmadd213ps %ymm4, %ymm6, %ymm5 - vaddps %ymm14, %ymm15, %ymm15 - vaddps %ymm15, %ymm13, %ymm15 - - /* - * Now compute an approximate reciprocal to mix into the computation - * To avoid any danger of nonportability, force it to 12 bits, - * though I suspect it always is anyway on current platforms. - */ - vrcpps %ymm9, %ymm13 - vandps %ymm1, %ymm13, %ymm13 - /* - * Now compute the error sEr where sRecip_hi = (1/R_hi) * (1 - sEr) - * so that we can compensate for it. - */ - vmovups _sOne+__svml_stan_data_internal(%rip), %ymm1 - vfnmadd213ps %ymm1, %ymm13, %ymm9 - - /* - * Get a better approximation to 1/sR_hi (not far short of an ulp) - * using a third-order polynomial approximation - */ - vmovaps %ymm13, %ymm14 - vfmadd213ps %ymm13, %ymm9, %ymm14 - vfmadd231ps %ymm9, %ymm9, %ymm1 - vmulps %ymm1, %ymm14, %ymm1 - - /* - * Multiply by sRecip_ok to make sR_lo relative to sR_hi - * Since sR_lo is shifted off by about 12 bits, this is accurate enough. - */ - vmulps %ymm1, %ymm15, %ymm14 - - /* - * Now create a low reciprocal using - * (Recip_hi + Er * Recip_ok) * (1 + sR_lo^2 - sR_lo) - * =~= Recip_hi + Recip_ok * (Er + sR_lo^2 - sR_lo) - */ - vsubps %ymm9, %ymm14, %ymm9 - vfmsub213ps %ymm9, %ymm14, %ymm14 - vmulps %ymm14, %ymm1, %ymm9 - vmovaps %ymm2, %ymm1 - vfmadd213ps %ymm3, %ymm6, %ymm1 - vsubps %ymm1, %ymm3, %ymm15 - vmovaps %ymm8, %ymm3 - vfmadd213ps %ymm1, %ymm13, %ymm3 - vfmadd213ps %ymm15, %ymm6, %ymm2 - vfmsub213ps %ymm3, %ymm8, %ymm13 - vfmadd213ps %ymm2, %ymm8, %ymm9 - vaddps %ymm13, %ymm1, %ymm2 + /* Higher polynomial terms + Stage 1 (with unlimited parallelism) + P3 = C1_lo + C2 * Z. */ + vmovq 32(%rdx, %rax, 8), %xmm4 + vpcmpgtd %ymm1, %ymm9, %ymm9 + vmovmskps %ymm9, %eax - /* Z2 = Z^2 */ - vmulps %ymm6, %ymm6, %ymm1 - vaddps %ymm2, %ymm9, %ymm8 + vmovq 32(%rdx, %rcx, 8), %xmm1 - /* - * Stage 2 (with unlimited parallelism) - * P6 = C1_lo + C2 * Z + C3 * Z^2 + C4 * Z^3 - */ - vfmadd213ps %ymm7, %ymm5, %ymm1 + vinsertf128 $1, 32(%rdx, %r10, 8), %ymm4, %ymm4 + vinsertf128 $1, 32(%rdx, %r8, 8), %ymm1, %ymm1 + vpunpckldq %ymm1, %ymm4, %ymm9 - /* P9 = trail(dominant part) + C0_lo */ - vaddps 32(%rsp), %ymm8, %ymm5 + vmovq 32(%rdx, %rdi, 8), %xmm1 + vmovq 32(%rdx, %rsi, 8), %xmm4 - /* Final accumulation of low part */ - vfmadd213ps %ymm5, %ymm6, %ymm1 + vinsertf128 $1, 32(%rdx, %r9, 8), %ymm1, %ymm1 + vinsertf128 $1, 32(%rdx, %r11, 8), %ymm4, %ymm4 + vpunpckldq %ymm4, %ymm1, %ymm1 - /* And now the very final summation */ - vaddps %ymm1, %ymm3, %ymm6 + vpunpckhqdq %ymm1, %ymm9, %ymm4 + vpunpcklqdq %ymm1, %ymm9, %ymm9 - /* - * The end of implementation (LA with huge args reduction) - * End of large arguments path (_HA_, _LA_ and _EP_) - */ - vxorps %ymm11, %ymm6, %ymm11 + vmovups COMMON_DATA(_Neg4096)(%rip), %ymm1 - /* Merge results from main and large paths: */ - vblendvps %ymm10, %ymm11, %ymm0, %ymm0 + vfmadd213ps %ymm9, %ymm6, %ymm4 - /* Return to main vector processing path */ - jmp L(AUX_BRANCH_RETURN) - # LOE r12 r13 r14 r15 eax ymm0 ymm12 -END(_ZGVdN8v_tanf_avx2) + vandps %ymm1, %ymm2, %ymm9 + vsubps %ymm9, %ymm2, %ymm2 - .section .rodata, "a" - .align 32 + /* P4 = C3 + C4 * Z. */ + vaddps %ymm2, %ymm0, %ymm0 + vaddps %ymm0, %ymm13, %ymm0 -.FLT_15: - .long 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 - .type .FLT_15, @object - .size .FLT_15, 32 - .align 32 + /* Now compute an approximate reciprocal to mix into the computation + To avoid any danger of nonportability, force it to 12 bits, + though I suspect it always is anyway on current platforms. */ + vrcpps %ymm9, %ymm13 + vandps %ymm1, %ymm13, %ymm13 -.FLT_16: - .long 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 - .type .FLT_16, @object - .size .FLT_16, 32 - .align 32 + /* Now compute the error sEr where sRecip_hi = (1/R_hi) * (1 - sEr) + so that we can compensate for it. */ + vfnmadd213ps %ymm15, %ymm13, %ymm9 + + /* Get a better approximation to 1/sR_hi (not far short of an ulp) + using a third-order polynomial approximation. */ + vmovaps %ymm13, %ymm2 + vfmadd231ps %ymm9, %ymm9, %ymm15 + vfmadd213ps %ymm13, %ymm9, %ymm2 + vmulps %ymm15, %ymm2, %ymm1 + + /* Multiply by sRecip_ok to make sR_lo relative to sR_hi Since + sR_lo is shifted off by about 12 bits, this is accurate + enough. */ + vmulps %ymm1, %ymm0, %ymm2 + + /* Now create a low reciprocal using + (Recip_hi + Er * Recip_ok) * (1 + sR_lo^2 - sR_lo) + =~= Recip_hi + Recip_ok * (Er + sR_lo^2 - sR_lo). */ + vsubps %ymm9, %ymm2, %ymm9 + vfmsub213ps %ymm9, %ymm2, %ymm2 + vmulps %ymm2, %ymm1, %ymm9 + vmovaps %ymm14, %ymm1 + vfmadd213ps %ymm3, %ymm6, %ymm1 + vsubps %ymm1, %ymm3, %ymm0 + vmovaps %ymm8, %ymm3 + vfmadd213ps %ymm1, %ymm13, %ymm3 + vfmadd213ps %ymm0, %ymm6, %ymm14 + vfmsub213ps %ymm3, %ymm8, %ymm13 + vfmadd213ps %ymm14, %ymm8, %ymm9 + vaddps %ymm13, %ymm1, %ymm2 -.FLT_17: - .long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff - .type .FLT_17, @object - .size .FLT_17, 32 - .align 32 + /* Z2 = Z^2. */ + vmulps %ymm6, %ymm6, %ymm1 + vaddps %ymm2, %ymm9, %ymm2 -.FLT_18: - .long 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000 - .type .FLT_18, @object - .size .FLT_18, 32 - .align 32 + /* Stage 2 (with unlimited parallelism) + P6 = C1_lo + C2 * Z + C3 * Z^2 + C4 * Z^3. */ + vfmadd213ps %ymm7, %ymm4, %ymm1 -.FLT_19: - .long 0x0000ffff, 0x0000ffff, 0x0000ffff, 0x0000ffff, 0x0000ffff, 0x0000ffff, 0x0000ffff, 0x0000ffff - .type .FLT_19, @object - .size .FLT_19, 32 - .align 32 + /* P9 = trail(dominant part) + C0_lo. */ + vaddps %ymm5, %ymm2, %ymm4 -.FLT_20: - .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 - .type .FLT_20, @object - .size .FLT_20, 32 - .align 32 + /* Final accumulation of low part. */ + vfmadd213ps %ymm4, %ymm6, %ymm1 -.FLT_21: - .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 - .type .FLT_21, @object - .size .FLT_21, 32 - .align 32 + /* And now the very final summation. */ + vaddps %ymm1, %ymm3, %ymm6 -.FLT_22: - .long 0x47400000, 0x47400000, 0x47400000, 0x47400000, 0x47400000, 0x47400000, 0x47400000, 0x47400000 - .type .FLT_22, @object - .size .FLT_22, 32 - .align 32 + /* The end of implementation (LA with huge args reduction) + End of large arguments path (_HA_, _LA_ and _EP_). */ + vpand %ymm6, %ymm10, %ymm6 + /* Merge results from main and large paths:. */ + vpxor %ymm6, %ymm12, %ymm0 -.FLT_23: - .long 0x28800000, 0x28800000, 0x28800000, 0x28800000, 0x28800000, 0x28800000, 0x28800000, 0x28800000 - .type .FLT_23, @object - .size .FLT_23, 32 - .align 32 -.FLT_24: - .long 0x0003ffff, 0x0003ffff, 0x0003ffff, 0x0003ffff, 0x0003ffff, 0x0003ffff, 0x0003ffff, 0x0003ffff - .type .FLT_24, @object - .size .FLT_24, 32 - .align 32 + /* `al` has 0 at special values. If all 1s `incb al` will + overflow and set zero flag. */ + incb %al + /* Go to special inputs processing branch. */ + jne L(SPECIAL_VALUES_BRANCH) + ret -.FLT_25: - .long 0x34000000, 0x34000000, 0x34000000, 0x34000000, 0x34000000, 0x34000000, 0x34000000, 0x34000000 - .type .FLT_25, @object - .size .FLT_25, 32 - .align 32 + /* Cold case. edx has 1s where there was a special value that + needs to be handled by a tanhf call. Optimize for code size + more so than speed here. */ +L(SPECIAL_VALUES_BRANCH): -.FLT_26: - .long 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff, 0x000001ff - .type .FLT_26, @object - .size .FLT_26, 32 - .align 32 + /* Use r13 to save/restore the stack. This allows us to use rbp + as callee save register saving code size. */ + pushq %r13 + cfi_adjust_cfa_offset (8) + cfi_offset (r13, -16) + /* Need to callee save registers to preserve state across tanhf + calls. */ + pushq %rbx + cfi_adjust_cfa_offset (8) + cfi_offset (rbx, -24) + pushq %rbp + cfi_adjust_cfa_offset (8) + cfi_offset (rbp, -32) + movq %rsp, %r13 + cfi_def_cfa_register (r13) -.FLT_27: - .long 0x40c90fdb, 0x40c90fdb, 0x40c90fdb, 0x40c90fdb, 0x40c90fdb, 0x40c90fdb, 0x40c90fdb, 0x40c90fdb - .type .FLT_27, @object - .size .FLT_27, 32 - .align 32 + /* Align stack and make room for 2x ymm vectors. */ + andq $-32, %rsp + addq $-64, %rsp -.FLT_28: - .long 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e, 0xb43bbd2e - .type .FLT_28, @object - .size .FLT_28, 32 - .align 32 -.FLT_29: - .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff - .type .FLT_29, @object - .size .FLT_29, 32 - .align 32 + /* Save original input (ymm0 unchanged up to this point). */ + vmovaps %ymm11, 32(%rsp) + vmovaps %ymm0, (%rsp) -.FLT_30: - .long 0x35800000, 0x35800000, 0x35800000, 0x35800000, 0x35800000, 0x35800000, 0x35800000, 0x35800000 - .type .FLT_30, @object - .size .FLT_30, 32 - .align 32 + vzeroupper -.FLT_31: - .long 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff - .type .FLT_31, @object - .size .FLT_31, 32 - .align 32 + /* eax has 1s where there was a special value that needs to be + handled by a tanf call. */ + negb %al + movzbl %al, %ebx +L(SPECIAL_VALUES_LOOP): -.FLT_32: - .long 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f, 0x0000007f - .type .FLT_32, @object - .size .FLT_32, 32 - .align 32 + /* use rbp as index for special value that is saved across calls + to tanhf. We technically don't need a callee save register + here as offset to rsp is always [0, 28] so we can restore + rsp by realigning to 64. Essentially the tradeoff is 1 extra + save/restore vs 2 extra instructions in the loop. Realigning + also costs more code size. */ + xorl %ebp, %ebp + tzcntl %ebx, %ebp + + /* Scalar math function call to process special input. */ + vmovss 32(%rsp, %rbp, 4), %xmm0 + call tanf@PLT -.FLT_33: - .long 0xfffff000, 0xfffff000, 0xfffff000, 0xfffff000, 0xfffff000, 0xfffff000, 0xfffff000, 0xfffff000 - .type .FLT_33, @object - .size .FLT_33, 32 - .align 32 + /* No good way to avoid the store-forwarding fault this will + cause on return. `lfence` avoids the SF fault but at greater + cost as it serialized stack/callee save restoration. */ + vmovss %xmm0, (%rsp, %rbp, 4) -#ifdef __svml_stan_data_internal_typedef -typedef unsigned int VUINT32; -typedef struct { - __declspec(align(32)) VUINT32 _sInvPI_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPI1_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPI2_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPI3_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPI2_ha_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPI3_ha_uisa[8][1]; - __declspec(align(32)) VUINT32 Th_tbl_uisa[32][1]; - __declspec(align(32)) VUINT32 Tl_tbl_uisa[32][1]; - __declspec(align(32)) VUINT32 _sPC3_uisa[8][1]; - __declspec(align(32)) VUINT32 _sPC5_uisa[8][1]; - __declspec(align(32)) VUINT32 _sRangeReductionVal_uisa[8][1]; - __declspec(align(32)) VUINT32 _sInvPi[8][1]; - __declspec(align(32)) VUINT32 _sSignMask[8][1]; - __declspec(align(32)) VUINT32 _sAbsMask[8][1]; - __declspec(align(32)) VUINT32 _sRangeVal[8][1]; - __declspec(align(32)) VUINT32 _sRShifter[8][1]; - __declspec(align(32)) VUINT32 _sOne[8][1]; - __declspec(align(32)) VUINT32 _sRangeReductionVal[8][1]; - __declspec(align(32)) VUINT32 _sPI1[8][1]; - __declspec(align(32)) VUINT32 _sPI2[8][1]; - __declspec(align(32)) VUINT32 _sPI3[8][1]; - __declspec(align(32)) VUINT32 _sPI4[8][1]; - __declspec(align(32)) VUINT32 _sPI1_FMA[8][1]; - __declspec(align(32)) VUINT32 _sPI2_FMA[8][1]; - __declspec(align(32)) VUINT32 _sPI3_FMA[8][1]; - __declspec(align(32)) VUINT32 _sP0[8][1]; - __declspec(align(32)) VUINT32 _sP1[8][1]; - __declspec(align(32)) VUINT32 _sQ0[8][1]; - __declspec(align(32)) VUINT32 _sQ1[8][1]; - __declspec(align(32)) VUINT32 _sQ2[8][1]; - __declspec(align(32)) VUINT32 _sTwo[8][1]; - __declspec(align(32)) VUINT32 _sCoeffs[128][10][1]; -} __svml_stan_data_internal; -#endif -__svml_stan_data_internal: - /* UISA */ - .long 0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983 /* _sInvPI_uisa */ - .align 32 - .long 0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda /* _sPI1_uisa */ - .align 32 - .long 0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168 /* _sPI2_uisa */ - .align 32 - .long 0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5 /* _sPI3_uisa */ - .align 32 - .long 0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000 /* _sPI2_ha_uisa */ - .align 32 - .long 0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a /* _sPI3_ha_uisa */ - /* Th_tbl_uisa for i from 0 to 31 do printsingle(tan(i*Pi/32)); */ - .align 32 - .long 0x80000000, 0x3dc9b5dc, 0x3e4bafaf, 0x3e9b5042 - .long 0x3ed413cd, 0x3f08d5b9, 0x3f2b0dc1, 0x3f521801 - .long 0x3f800000, 0x3f9bf7ec, 0x3fbf90c7, 0x3fef789e - .long 0x401a827a, 0x4052facf, 0x40a0dff7, 0x41227363 - .long 0xff7fffff, 0xc1227363, 0xc0a0dff7, 0xc052facf - .long 0xc01a827a, 0xbfef789e, 0xbfbf90c7, 0xbf9bf7ec - .long 0xbf800000, 0xbf521801, 0xbf2b0dc1, 0xbf08d5b9 - .long 0xbed413cd, 0xbe9b5042, 0xbe4bafaf, 0xbdc9b5dc - /* Tl_tbl_uisa for i from 0 to 31 do printsingle(tan(i*Pi/32)-round(tan(i*Pi/32), SG, RN)); */ - .align 32 - .long 0x80000000, 0x3145b2da, 0x2f2a62b0, 0xb22a39c2 - .long 0xb1c0621a, 0xb25ef963, 0x32ab7f99, 0x32ae4285 - .long 0x00000000, 0x33587608, 0x32169d18, 0xb30c3ec0 - .long 0xb3cc0622, 0x3390600e, 0x331091dc, 0xb454a046 - .long 0xf3800000, 0x3454a046, 0xb31091dc, 0xb390600e - .long 0x33cc0622, 0x330c3ec0, 0xb2169d18, 0xb3587608 - .long 0x00000000, 0xb2ae4285, 0xb2ab7f99, 0x325ef963 - .long 0x31c0621a, 0x322a39c2, 0xaf2a62b0, 0xb145b2da - .align 32 - .long 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6 /* _sPC3_uisa */ - .align 32 - .long 0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888 /* _sPC5_uisa */ - .align 32 - .long 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000 /* _sRangeReductionVal_uisa */ - .align 32 - .long 0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983 /* _sInvPi */ - .align 32 - .long 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000, 0x80000000 /* _sSignMask */ - .align 32 - .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF /* _sAbsMask */ - .align 32 - .long 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 /* _sRangeVal */ - .align 32 - .long 0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000 /* _sRShifter */ - .align 32 - .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 /* _sOne */ - .align 32 - .long 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000, 0x46010000 /* _sRangeVal */ - .align 32 - .long 0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000 /* _sPI1 */ - .align 32 - .long 0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000 /* _sPI2 */ - .align 32 - .long 0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000 /* _sPI3 */ - .align 32 - .long 0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A /* _sPI4 */ - // PI1, PI2, and PI3 when FMA is available - .align 32 - .long 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB /* _sPI1_FMA */ - .align 32 - .long 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E /* _sPI2_FMA */ - .align 32 - .long 0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED /* _sPI3_FMA */ - .align 32 - .long 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC /* _sP0 */ - .align 32 - .long 0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4 /* _sP1 */ - .align 32 - .long 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC /* _sQ0 */ - .align 32 - .long 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB /* _sQ1 */ - .align 32 - .long 0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B /* _sQ2 */ - .align 32 - .long 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000 /* _sTwo */ - // _sCoeffs Breakpoint B = 0 * pi/128, function tan(B + x) - .align 32 - .long 0x3FC90FDB // B' = pi/2 - B (high single) - .long 0xB33BBD2E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x00000000 // c0 (high single) - .long 0x00000000 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x00000000 // c1 (low single) - .long 0x00000000 // c2 - .long 0x3EAAACDD // c3 - .long 0x00000000 // c4 - .long 0x3FC5EB9B // B' = pi/2 - B (high single) - .long 0x32DE638C // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3CC91A31 // c0 (high single) - .long 0x2F8E8D1A // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3A1DFA00 // c1 (low single) - .long 0x3CC9392D // c2 - .long 0x3EAB1889 // c3 - .long 0x3C885D3B // c4 - .long 0x3FC2C75C // B' = pi/2 - B (high single) - .long 0xB2CBBE8A // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3D49393C // c0 (high single) - .long 0x30A39F5B // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3B1E2B00 // c1 (low single) - .long 0x3D49B5D4 // c2 - .long 0x3EAC4F10 // c3 - .long 0x3CFD9425 // c4 - .long 0x3FBFA31C // B' = pi/2 - B (high single) - .long 0x33450FB0 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3D9711CE // c0 (high single) - .long 0x314FEB28 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3BB24C00 // c1 (low single) - .long 0x3D97E43A // c2 - .long 0x3EAE6A89 // c3 - .long 0x3D4D07E0 // c4 - .long 0x3FBC7EDD // B' = pi/2 - B (high single) - .long 0xB1800ADD // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3DC9B5DC // c0 (high single) - .long 0x3145AD86 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3C1EEF20 // c1 (low single) - .long 0x3DCBAAEA // c2 - .long 0x3EB14E5E // c3 - .long 0x3D858BB2 // c4 - .long 0x3FB95A9E // B' = pi/2 - B (high single) - .long 0xB3651267 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3DFC98C2 // c0 (high single) - .long 0xB0AE525C // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3C793D20 // c1 (low single) - .long 0x3E003845 // c2 - .long 0x3EB5271F // c3 - .long 0x3DAC669E // c4 - .long 0x3FB6365E // B' = pi/2 - B (high single) - .long 0x328BB91C // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E17E564 // c0 (high single) - .long 0xB1C5A2E4 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3CB440D0 // c1 (low single) - .long 0x3E1B3D00 // c2 - .long 0x3EB9F664 // c3 - .long 0x3DD647C0 // c4 - .long 0x3FB3121F // B' = pi/2 - B (high single) - .long 0xB30F347D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E31AE4D // c0 (high single) - .long 0xB1F32251 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3CF6A500 // c1 (low single) - .long 0x3E3707DA // c2 - .long 0x3EBFA489 // c3 - .long 0x3DFBD9C7 // c4 - .long 0x3FAFEDDF // B' = pi/2 - B (high single) - .long 0x331BBA77 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E4BAFAF // c0 (high single) - .long 0x2F2A29E0 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D221018 // c1 (low single) - .long 0x3E53BED0 // c2 - .long 0x3EC67E26 // c3 - .long 0x3E1568E2 // c4 - .long 0x3FACC9A0 // B' = pi/2 - B (high single) - .long 0xB2655A50 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E65F267 // c0 (high single) - .long 0x31B4B1DF // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D4E8B90 // c1 (low single) - .long 0x3E718ACA // c2 - .long 0x3ECE7164 // c3 - .long 0x3E2DC161 // c4 - .long 0x3FA9A560 // B' = pi/2 - B (high single) - .long 0x33719861 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E803FD4 // c0 (high single) - .long 0xB2279E66 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D807FC8 // c1 (low single) - .long 0x3E884BD4 // c2 - .long 0x3ED7812D // c3 - .long 0x3E4636EB // c4 - .long 0x3FA68121 // B' = pi/2 - B (high single) - .long 0x31E43AAC // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E8DB082 // c0 (high single) - .long 0xB132A234 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D9CD7D0 // c1 (low single) - .long 0x3E988A60 // c2 - .long 0x3EE203E3 // c3 - .long 0x3E63582C // c4 - .long 0x3FA35CE2 // B' = pi/2 - B (high single) - .long 0xB33889B6 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3E9B5042 // c0 (high single) - .long 0xB22A3AEE // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3DBC7490 // c1 (low single) - .long 0x3EA99AF5 // c2 - .long 0x3EEDE107 // c3 - .long 0x3E80E9AA // c4 - .long 0x3FA038A2 // B' = pi/2 - B (high single) - .long 0x32E4CA7E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3EA92457 // c0 (high single) - .long 0x30B80830 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3DDF8200 // c1 (low single) - .long 0x3EBB99E9 // c2 - .long 0x3EFB4AA8 // c3 - .long 0x3E9182BE // c4 - .long 0x3F9D1463 // B' = pi/2 - B (high single) - .long 0xB2C55799 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3EB73250 // c0 (high single) - .long 0xB2028823 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E0318F8 // c1 (low single) - .long 0x3ECEA678 // c2 - .long 0x3F053C67 // c3 - .long 0x3EA41E53 // c4 - .long 0x3F99F023 // B' = pi/2 - B (high single) - .long 0x33484328 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3EC5800D // c0 (high single) - .long 0xB214C3C1 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E185E54 // c1 (low single) - .long 0x3EE2E342 // c2 - .long 0x3F0DCA73 // c3 - .long 0x3EB8CC21 // c4 - .long 0x3F96CBE4 // B' = pi/2 - B (high single) - .long 0xB14CDE2E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3ED413CD // c0 (high single) - .long 0xB1C06152 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E2FB0CC // c1 (low single) - .long 0x3EF876CB // c2 - .long 0x3F177807 // c3 - .long 0x3ED08437 // c4 - .long 0x3F93A7A5 // B' = pi/2 - B (high single) - .long 0xB361DEEE // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3EE2F439 // c0 (high single) - .long 0xB1F4399E // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E49341C // c1 (low single) - .long 0x3F07C61A // c2 - .long 0x3F22560F // c3 - .long 0x3EEAA81E // c4 - .long 0x3F908365 // B' = pi/2 - B (high single) - .long 0x3292200D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3EF22870 // c0 (high single) - .long 0x325271F4 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E65107A // c1 (low single) - .long 0x3F1429F0 // c2 - .long 0x3F2E8AFC // c3 - .long 0x3F040498 // c4 - .long 0x3F8D5F26 // B' = pi/2 - B (high single) - .long 0xB30C0105 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F00DC0D // c0 (high single) - .long 0xB214AF72 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E81B994 // c1 (low single) - .long 0x3F218233 // c2 - .long 0x3F3C4531 // c3 - .long 0x3F149688 // c4 - .long 0x3F8A3AE6 // B' = pi/2 - B (high single) - .long 0x331EEDF0 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F08D5B9 // c0 (high single) - .long 0xB25EF98E // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E92478D // c1 (low single) - .long 0x3F2FEDC9 // c2 - .long 0x3F4BCD58 // c3 - .long 0x3F27AE9E // c4 - .long 0x3F8716A7 // B' = pi/2 - B (high single) - .long 0xB2588C6D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F1105AF // c0 (high single) - .long 0x32F045B0 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EA44EE2 // c1 (low single) - .long 0x3F3F8FDB // c2 - .long 0x3F5D3FD0 // c3 - .long 0x3F3D0A23 // c4 - .long 0x3F83F267 // B' = pi/2 - B (high single) - .long 0x3374CBD9 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F1970C4 // c0 (high single) - .long 0x32904848 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EB7EFF8 // c1 (low single) - .long 0x3F50907C // c2 - .long 0x3F710FEA // c3 - .long 0x3F561FED // c4 - .long 0x3F80CE28 // B' = pi/2 - B (high single) - .long 0x31FDD672 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F221C37 // c0 (high single) - .long 0xB20C61DC // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3ECD4F71 // c1 (low single) - .long 0x3F631DAA // c2 - .long 0x3F83B471 // c3 - .long 0x3F7281EA // c4 - .long 0x3F7B53D1 // B' = pi/2 - B (high single) - .long 0x32955386 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F2B0DC1 // c0 (high single) - .long 0x32AB7EBA // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EE496C2 // c1 (low single) - .long 0x3F776C40 // c2 - .long 0x3F9065C1 // c3 - .long 0x3F89AFB6 // c4 - .long 0x3F750B52 // B' = pi/2 - B (high single) - .long 0x32EB316F // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F344BA9 // c0 (high single) - .long 0xB2B8B0EA // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EFDF4F7 // c1 (low single) - .long 0x3F86DCA8 // c2 - .long 0x3F9ED53B // c3 - .long 0x3F9CBEDE // c4 - .long 0x3F6EC2D4 // B' = pi/2 - B (high single) - .long 0xB2BEF0A7 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F3DDCCF // c0 (high single) - .long 0x32D29606 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEE6606F // c1 (low single) - .long 0x3F9325D6 // c2 - .long 0x3FAF4E69 // c3 - .long 0x3FB3080C // c4 - .long 0x3F687A55 // B' = pi/2 - B (high single) - .long 0xB252257B // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F47C8CC // c0 (high single) - .long 0xB200F51A // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEC82C6C // c1 (low single) - .long 0x3FA0BAE9 // c2 - .long 0x3FC2252F // c3 - .long 0x3FCD24C7 // c4 - .long 0x3F6231D6 // B' = pi/2 - B (high single) - .long 0xB119A6A2 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F521801 // c0 (high single) - .long 0x32AE4178 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEA72938 // c1 (low single) - .long 0x3FAFCC22 // c2 - .long 0x3FD7BD4A // c3 - .long 0x3FEBB01B // c4 - .long 0x3F5BE957 // B' = pi/2 - B (high single) - .long 0x3205522A // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F5CD3BE // c0 (high single) - .long 0x31460308 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBE8306C5 // c1 (low single) - .long 0x3FC09232 // c2 - .long 0x3FF09632 // c3 - .long 0x4007DB00 // c4 - .long 0x3F55A0D8 // B' = pi/2 - B (high single) - .long 0x329886FF // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F68065E // c0 (high single) - .long 0x32670D1A // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBE36D1D6 // c1 (low single) - .long 0x3FD35007 // c2 - .long 0x4006A861 // c3 - .long 0x401D4BDA // c4 - .long 0x3F4F5859 // B' = pi/2 - B (high single) - .long 0x32EE64E8 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0x3F73BB75 // c0 (high single) - .long 0x32FC908D // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBDBF94B0 // c1 (low single) - .long 0x3FE8550F // c2 - .long 0x40174F67 // c3 - .long 0x4036C608 // c4 - .long 0x3F490FDB // B' = pi/2 - B (high single) - .long 0xB2BBBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE8BE60E // c0 (high single) - .long 0x320D8D84 // c0 (low single) - .long 0x3F000000 // c1 (high 1 bit) - .long 0xBDF817B1 // c1 (low single) - .long 0xBD8345EB // c2 - .long 0x3D1DFDAC // c3 - .long 0xBC52CF6F // c4 - .long 0x3F42C75C // B' = pi/2 - B (high single) - .long 0xB24BBE8A // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE87283F // c0 (high single) - .long 0xB268B966 // c0 (low single) - .long 0x3F000000 // c1 (high 1 bit) - .long 0xBDFE6529 // c1 (low single) - .long 0xBD7B1953 // c2 - .long 0x3D18E109 // c3 - .long 0xBC4570B0 // c4 - .long 0x3F3C7EDD // B' = pi/2 - B (high single) - .long 0xB1000ADD // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE827420 // c0 (high single) - .long 0x320B8B4D // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DFB9428 // c1 (low single) - .long 0xBD7002B4 // c2 - .long 0x3D142A6C // c3 - .long 0xBC3A47FF // c4 - .long 0x3F36365E // B' = pi/2 - B (high single) - .long 0x320BB91C // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE7B9282 // c0 (high single) - .long 0xB13383D2 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DF5D211 // c1 (low single) - .long 0xBD6542B3 // c2 - .long 0x3D0FE5E5 // c3 - .long 0xBC31FB14 // c4 - .long 0x3F2FEDDF // B' = pi/2 - B (high single) - .long 0x329BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE724E73 // c0 (high single) - .long 0x3120C3E2 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DF05283 // c1 (low single) - .long 0xBD5AD45E // c2 - .long 0x3D0BAFBF // c3 - .long 0xBC27B8BB // c4 - .long 0x3F29A560 // B' = pi/2 - B (high single) - .long 0x32F19861 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE691B44 // c0 (high single) - .long 0x31F18936 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DEB138B // c1 (low single) - .long 0xBD50B2F7 // c2 - .long 0x3D07BE3A // c3 - .long 0xBC1E46A7 // c4 - .long 0x3F235CE2 // B' = pi/2 - B (high single) - .long 0xB2B889B6 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE5FF82C // c0 (high single) - .long 0xB170723A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DE61354 // c1 (low single) - .long 0xBD46DA06 // c2 - .long 0x3D0401F8 // c3 - .long 0xBC14E013 // c4 - .long 0x3F1D1463 // B' = pi/2 - B (high single) - .long 0xB2455799 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE56E46B // c0 (high single) - .long 0x31E3F001 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DE15025 // c1 (low single) - .long 0xBD3D4550 // c2 - .long 0x3D00462D // c3 - .long 0xBC092C98 // c4 - .long 0x3F16CBE4 // B' = pi/2 - B (high single) - .long 0xB0CCDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE4DDF41 // c0 (high single) - .long 0xB1AEA094 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DDCC85C // c1 (low single) - .long 0xBD33F0BE // c2 - .long 0x3CFA23B0 // c3 - .long 0xBC01FCF7 // c4 - .long 0x3F108365 // B' = pi/2 - B (high single) - .long 0x3212200D // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE44E7F8 // c0 (high single) - .long 0xB1CAA3CB // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD87A74 // c1 (low single) - .long 0xBD2AD885 // c2 - .long 0x3CF3C785 // c3 - .long 0xBBF1E348 // c4 - .long 0x3F0A3AE6 // B' = pi/2 - B (high single) - .long 0x329EEDF0 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE3BFDDC // c0 (high single) - .long 0xB132521A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD464FC // c1 (low single) - .long 0xBD21F8F1 // c2 - .long 0x3CEE3076 // c3 - .long 0xBBE6D263 // c4 - .long 0x3F03F267 // B' = pi/2 - B (high single) - .long 0x32F4CBD9 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE33203E // c0 (high single) - .long 0x31FEF5BE // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD0869C // c1 (low single) - .long 0xBD194E8C // c2 - .long 0x3CE8DCA9 // c3 - .long 0xBBDADA55 // c4 - .long 0x3EFB53D1 // B' = pi/2 - B (high single) - .long 0x32155386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE2A4E71 // c0 (high single) - .long 0xB19CFCEC // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DCCDE11 // c1 (low single) - .long 0xBD10D605 // c2 - .long 0x3CE382A7 // c3 - .long 0xBBC8BD97 // c4 - .long 0x3EEEC2D4 // B' = pi/2 - B (high single) - .long 0xB23EF0A7 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE2187D0 // c0 (high single) - .long 0xB1B7C7F7 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC96A2B // c1 (low single) - .long 0xBD088C22 // c2 - .long 0x3CDE950E // c3 - .long 0xBBB89AD1 // c4 - .long 0x3EE231D6 // B' = pi/2 - B (high single) - .long 0xB099A6A2 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE18CBB7 // c0 (high single) - .long 0xAFE28430 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC629CE // c1 (low single) - .long 0xBD006DCD // c2 - .long 0x3CDA5A2C // c3 - .long 0xBBB0B3D2 // c4 - .long 0x3ED5A0D8 // B' = pi/2 - B (high single) - .long 0x321886FF // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE101985 // c0 (high single) - .long 0xB02FB2B8 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC31BF3 // c1 (low single) - .long 0xBCF0F04D // c2 - .long 0x3CD60BC7 // c3 - .long 0xBBA138BA // c4 - .long 0x3EC90FDB // B' = pi/2 - B (high single) - .long 0xB23BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBE07709D // c0 (high single) - .long 0xB18A2A83 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC03FA2 // c1 (low single) - .long 0xBCE15096 // c2 - .long 0x3CD26472 // c3 - .long 0xBB9A1270 // c4 - .long 0x3EBC7EDD // B' = pi/2 - B (high single) - .long 0xB0800ADD // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDFDA0CB // c0 (high single) - .long 0x2F14FCA0 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DBD93F7 // c1 (low single) - .long 0xBCD1F71B // c2 - .long 0x3CCEDD2B // c3 - .long 0xBB905946 // c4 - .long 0x3EAFEDDF // B' = pi/2 - B (high single) - .long 0x321BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDEC708C // c0 (high single) - .long 0xB14895C4 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DBB181E // c1 (low single) - .long 0xBCC2DEA6 // c2 - .long 0x3CCB5027 // c3 - .long 0xBB7F3969 // c4 - .long 0x3EA35CE2 // B' = pi/2 - B (high single) - .long 0xB23889B6 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDDB4F55 // c0 (high single) - .long 0x30F6437E // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB8CB52 // c1 (low single) - .long 0xBCB40210 // c2 - .long 0x3CC82D45 // c3 - .long 0xBB643075 // c4 - .long 0x3E96CBE4 // B' = pi/2 - B (high single) - .long 0xB04CDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDCA3BFF // c0 (high single) - .long 0x311C95EA // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB6ACDE // c1 (low single) - .long 0xBCA55C5B // c2 - .long 0x3CC5BC04 // c3 - .long 0xBB63A969 // c4 - .long 0x3E8A3AE6 // B' = pi/2 - B (high single) - .long 0x321EEDF0 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDB93569 // c0 (high single) - .long 0xAFB9ED00 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB4BC1F // c1 (low single) - .long 0xBC96E905 // c2 - .long 0x3CC2E6F5 // c3 - .long 0xBB3E10A6 // c4 - .long 0x3E7B53D1 // B' = pi/2 - B (high single) - .long 0x31955386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBDA83A77 // c0 (high single) - .long 0x316D967A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB2F87C // c1 (low single) - .long 0xBC88A31F // c2 - .long 0x3CC0E763 // c3 - .long 0xBB3F1666 // c4 - .long 0x3E6231D6 // B' = pi/2 - B (high single) - .long 0xB019A6A2 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD974A0D // c0 (high single) - .long 0xB14F365B // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB1616F // c1 (low single) - .long 0xBC750CD8 // c2 - .long 0x3CBEB595 // c3 - .long 0xBB22B883 // c4 - .long 0x3E490FDB // B' = pi/2 - B (high single) - .long 0xB1BBBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD866317 // c0 (high single) - .long 0xAFF02140 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAFF67D // c1 (low single) - .long 0xBC591CD0 // c2 - .long 0x3CBCBEAD // c3 - .long 0xBB04BBEC // c4 - .long 0x3E2FEDDF // B' = pi/2 - B (high single) - .long 0x319BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD6B08FF // c0 (high single) - .long 0xB0EED236 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAEB739 // c1 (low single) - .long 0xBC3D6D51 // c2 - .long 0x3CBB485D // c3 - .long 0xBAFFF5BA // c4 - .long 0x3E16CBE4 // B' = pi/2 - B (high single) - .long 0xAFCCDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD495A6C // c0 (high single) - .long 0xB0A427BD // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DADA345 // c1 (low single) - .long 0xBC21F648 // c2 - .long 0x3CB9D1B4 // c3 - .long 0xBACB5567 // c4 - .long 0x3DFB53D1 // B' = pi/2 - B (high single) - .long 0x31155386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD27B856 // c0 (high single) - .long 0xB0F7EE91 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DACBA4E // c1 (low single) - .long 0xBC06AEE3 // c2 - .long 0x3CB8E5DC // c3 - .long 0xBAEC00EE // c4 - .long 0x3DC90FDB // B' = pi/2 - B (high single) - .long 0xB13BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBD0620A3 // c0 (high single) - .long 0xB0ECAB40 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DABFC11 // c1 (low single) - .long 0xBBD7200F // c2 - .long 0x3CB79475 // c3 - .long 0xBA2B0ADC // c4 - .long 0x3D96CBE4 // B' = pi/2 - B (high single) - .long 0xAF4CDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBCC92278 // c0 (high single) - .long 0x302F2E68 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAB6854 // c1 (low single) - .long 0xBBA1214F // c2 - .long 0x3CB6C1E9 // c3 - .long 0x3843C2F3 // c4 - .long 0x3D490FDB // B' = pi/2 - B (high single) - .long 0xB0BBBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBC861015 // c0 (high single) - .long 0xAFD68E2E // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAAFEEB // c1 (low single) - .long 0xBB569F3F // c2 - .long 0x3CB6A84E // c3 - .long 0xBAC64194 // c4 - .long 0x3CC90FDB // B' = pi/2 - B (high single) - .long 0xB03BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0xBC060BF3 // c0 (high single) - .long 0x2FE251AE // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAABFB9 // c1 (low single) - .long 0xBAD67C60 // c2 - .long 0x3CB64CA5 // c3 - .long 0xBACDE881 // c4 - .long 0x00000000 // B' = pi/2 - B (high single) - .long 0x00000000 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x00000000 // c0 (high single) - .long 0x00000000 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAAAAAB // c1 (low single) - .long 0x00000000 // c2 - .long 0x3CB5E28B // c3 - .long 0x00000000 // c4 - .long 0xBCC90FDB // B' = pi/2 - B (high single) - .long 0x303BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3C060BF3 // c0 (high single) - .long 0xAFE251AE // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAABFB9 // c1 (low single) - .long 0x3AD67C60 // c2 - .long 0x3CB64CA5 // c3 - .long 0x3ACDE881 // c4 - .long 0xBD490FDB // B' = pi/2 - B (high single) - .long 0x30BBBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3C861015 // c0 (high single) - .long 0x2FD68E2E // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAAFEEB // c1 (low single) - .long 0x3B569F3F // c2 - .long 0x3CB6A84E // c3 - .long 0x3AC64194 // c4 - .long 0xBD96CBE4 // B' = pi/2 - B (high single) - .long 0x2F4CDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3CC92278 // c0 (high single) - .long 0xB02F2E68 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAB6854 // c1 (low single) - .long 0x3BA1214F // c2 - .long 0x3CB6C1E9 // c3 - .long 0xB843C2F2 // c4 - .long 0xBDC90FDB // B' = pi/2 - B (high single) - .long 0x313BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D0620A3 // c0 (high single) - .long 0x30ECAB40 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DABFC11 // c1 (low single) - .long 0x3BD7200F // c2 - .long 0x3CB79475 // c3 - .long 0x3A2B0ADC // c4 - .long 0xBDFB53D1 // B' = pi/2 - B (high single) - .long 0xB1155386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D27B856 // c0 (high single) - .long 0x30F7EE91 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DACBA4E // c1 (low single) - .long 0x3C06AEE3 // c2 - .long 0x3CB8E5DC // c3 - .long 0x3AEC00EE // c4 - .long 0xBE16CBE4 // B' = pi/2 - B (high single) - .long 0x2FCCDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D495A6C // c0 (high single) - .long 0x30A427BD // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DADA345 // c1 (low single) - .long 0x3C21F648 // c2 - .long 0x3CB9D1B4 // c3 - .long 0x3ACB5567 // c4 - .long 0xBE2FEDDF // B' = pi/2 - B (high single) - .long 0xB19BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D6B08FF // c0 (high single) - .long 0x30EED236 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAEB739 // c1 (low single) - .long 0x3C3D6D51 // c2 - .long 0x3CBB485D // c3 - .long 0x3AFFF5BA // c4 - .long 0xBE490FDB // B' = pi/2 - B (high single) - .long 0x31BBBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D866317 // c0 (high single) - .long 0x2FF02140 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DAFF67D // c1 (low single) - .long 0x3C591CD0 // c2 - .long 0x3CBCBEAD // c3 - .long 0x3B04BBEC // c4 - .long 0xBE6231D6 // B' = pi/2 - B (high single) - .long 0x3019A6A2 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3D974A0D // c0 (high single) - .long 0x314F365B // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB1616F // c1 (low single) - .long 0x3C750CD8 // c2 - .long 0x3CBEB595 // c3 - .long 0x3B22B883 // c4 - .long 0xBE7B53D1 // B' = pi/2 - B (high single) - .long 0xB1955386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DA83A77 // c0 (high single) - .long 0xB16D967A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB2F87C // c1 (low single) - .long 0x3C88A31F // c2 - .long 0x3CC0E763 // c3 - .long 0x3B3F1666 // c4 - .long 0xBE8A3AE6 // B' = pi/2 - B (high single) - .long 0xB21EEDF0 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DB93569 // c0 (high single) - .long 0x2FB9ED00 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB4BC1F // c1 (low single) - .long 0x3C96E905 // c2 - .long 0x3CC2E6F5 // c3 - .long 0x3B3E10A6 // c4 - .long 0xBE96CBE4 // B' = pi/2 - B (high single) - .long 0x304CDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DCA3BFF // c0 (high single) - .long 0xB11C95EA // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB6ACDE // c1 (low single) - .long 0x3CA55C5B // c2 - .long 0x3CC5BC04 // c3 - .long 0x3B63A969 // c4 - .long 0xBEA35CE2 // B' = pi/2 - B (high single) - .long 0x323889B6 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DDB4F55 // c0 (high single) - .long 0xB0F6437E // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DB8CB52 // c1 (low single) - .long 0x3CB40210 // c2 - .long 0x3CC82D45 // c3 - .long 0x3B643075 // c4 - .long 0xBEAFEDDF // B' = pi/2 - B (high single) - .long 0xB21BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DEC708C // c0 (high single) - .long 0x314895C4 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DBB181E // c1 (low single) - .long 0x3CC2DEA6 // c2 - .long 0x3CCB5027 // c3 - .long 0x3B7F3969 // c4 - .long 0xBEBC7EDD // B' = pi/2 - B (high single) - .long 0x30800ADD // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3DFDA0CB // c0 (high single) - .long 0xAF14FCA0 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DBD93F7 // c1 (low single) - .long 0x3CD1F71B // c2 - .long 0x3CCEDD2B // c3 - .long 0x3B905946 // c4 - .long 0xBEC90FDB // B' = pi/2 - B (high single) - .long 0x323BBD2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E07709D // c0 (high single) - .long 0x318A2A83 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC03FA2 // c1 (low single) - .long 0x3CE15096 // c2 - .long 0x3CD26472 // c3 - .long 0x3B9A1270 // c4 - .long 0xBED5A0D8 // B' = pi/2 - B (high single) - .long 0xB21886FF // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E101985 // c0 (high single) - .long 0x302FB2B8 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC31BF3 // c1 (low single) - .long 0x3CF0F04D // c2 - .long 0x3CD60BC7 // c3 - .long 0x3BA138BA // c4 - .long 0xBEE231D6 // B' = pi/2 - B (high single) - .long 0x3099A6A2 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E18CBB7 // c0 (high single) - .long 0x2FE28430 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC629CE // c1 (low single) - .long 0x3D006DCD // c2 - .long 0x3CDA5A2C // c3 - .long 0x3BB0B3D2 // c4 - .long 0xBEEEC2D4 // B' = pi/2 - B (high single) - .long 0x323EF0A7 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E2187D0 // c0 (high single) - .long 0x31B7C7F7 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DC96A2B // c1 (low single) - .long 0x3D088C22 // c2 - .long 0x3CDE950E // c3 - .long 0x3BB89AD1 // c4 - .long 0xBEFB53D1 // B' = pi/2 - B (high single) - .long 0xB2155386 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E2A4E71 // c0 (high single) - .long 0x319CFCEC // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DCCDE11 // c1 (low single) - .long 0x3D10D605 // c2 - .long 0x3CE382A7 // c3 - .long 0x3BC8BD97 // c4 - .long 0xBF03F267 // B' = pi/2 - B (high single) - .long 0xB2F4CBD9 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E33203E // c0 (high single) - .long 0xB1FEF5BE // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD0869C // c1 (low single) - .long 0x3D194E8C // c2 - .long 0x3CE8DCA9 // c3 - .long 0x3BDADA55 // c4 - .long 0xBF0A3AE6 // B' = pi/2 - B (high single) - .long 0xB29EEDF0 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E3BFDDC // c0 (high single) - .long 0x3132521A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD464FC // c1 (low single) - .long 0x3D21F8F1 // c2 - .long 0x3CEE3076 // c3 - .long 0x3BE6D263 // c4 - .long 0xBF108365 // B' = pi/2 - B (high single) - .long 0xB212200D // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E44E7F8 // c0 (high single) - .long 0x31CAA3CB // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DD87A74 // c1 (low single) - .long 0x3D2AD885 // c2 - .long 0x3CF3C785 // c3 - .long 0x3BF1E348 // c4 - .long 0xBF16CBE4 // B' = pi/2 - B (high single) - .long 0x30CCDE2E // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E4DDF41 // c0 (high single) - .long 0x31AEA094 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DDCC85C // c1 (low single) - .long 0x3D33F0BE // c2 - .long 0x3CFA23B0 // c3 - .long 0x3C01FCF7 // c4 - .long 0xBF1D1463 // B' = pi/2 - B (high single) - .long 0x32455799 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E56E46B // c0 (high single) - .long 0xB1E3F001 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DE15025 // c1 (low single) - .long 0x3D3D4550 // c2 - .long 0x3D00462D // c3 - .long 0x3C092C98 // c4 - .long 0xBF235CE2 // B' = pi/2 - B (high single) - .long 0x32B889B6 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E5FF82C // c0 (high single) - .long 0x3170723A // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DE61354 // c1 (low single) - .long 0x3D46DA06 // c2 - .long 0x3D0401F8 // c3 - .long 0x3C14E013 // c4 - .long 0xBF29A560 // B' = pi/2 - B (high single) - .long 0xB2F19861 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E691B44 // c0 (high single) - .long 0xB1F18936 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DEB138B // c1 (low single) - .long 0x3D50B2F7 // c2 - .long 0x3D07BE3A // c3 - .long 0x3C1E46A7 // c4 - .long 0xBF2FEDDF // B' = pi/2 - B (high single) - .long 0xB29BBA77 // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E724E73 // c0 (high single) - .long 0xB120C3E2 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DF05283 // c1 (low single) - .long 0x3D5AD45E // c2 - .long 0x3D0BAFBF // c3 - .long 0x3C27B8BB // c4 - .long 0xBF36365E // B' = pi/2 - B (high single) - .long 0xB20BB91C // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E7B9282 // c0 (high single) - .long 0x313383D2 // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DF5D211 // c1 (low single) - .long 0x3D6542B3 // c2 - .long 0x3D0FE5E5 // c3 - .long 0x3C31FB14 // c4 - .long 0xBF3C7EDD // B' = pi/2 - B (high single) - .long 0x31000ADD // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E827420 // c0 (high single) - .long 0xB20B8B4D // c0 (low single) - .long 0x3E800000 // c1 (high 1 bit) - .long 0x3DFB9428 // c1 (low single) - .long 0x3D7002B4 // c2 - .long 0x3D142A6C // c3 - .long 0x3C3A47FF // c4 - .long 0xBF42C75C // B' = pi/2 - B (high single) - .long 0x324BBE8A // B' = pi/2 - B (low single) - .long 0x3F800000 // tau (1 for cot path) - .long 0x3E87283F // c0 (high single) - .long 0x3268B966 // c0 (low single) - .long 0x3F000000 // c1 (high 1 bit) - .long 0xBDFE6529 // c1 (low single) - .long 0x3D7B1953 // c2 - .long 0x3D18E109 // c3 - .long 0x3C4570B0 // c4 - .long 0xBF490FDB // B' = pi/2 - B (high single) - .long 0x32BBBD2E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF800000 // c0 (high single) - .long 0x2B410000 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xB3000000 // c1 (low single) - .long 0xC0000000 // c2 - .long 0x402AB7C8 // c3 - .long 0xC05561DB // c4 - .long 0xBF4F5859 // B' = pi/2 - B (high single) - .long 0xB2EE64E8 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF73BB75 // c0 (high single) - .long 0xB2FC908D // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBDBF94B0 // c1 (low single) - .long 0xBFE8550F // c2 - .long 0x40174F67 // c3 - .long 0xC036C608 // c4 - .long 0xBF55A0D8 // B' = pi/2 - B (high single) - .long 0xB29886FF // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF68065E // c0 (high single) - .long 0xB2670D1A // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBE36D1D6 // c1 (low single) - .long 0xBFD35007 // c2 - .long 0x4006A861 // c3 - .long 0xC01D4BDA // c4 - .long 0xBF5BE957 // B' = pi/2 - B (high single) - .long 0xB205522A // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF5CD3BE // c0 (high single) - .long 0xB1460308 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBE8306C5 // c1 (low single) - .long 0xBFC09232 // c2 - .long 0x3FF09632 // c3 - .long 0xC007DB00 // c4 - .long 0xBF6231D6 // B' = pi/2 - B (high single) - .long 0x3119A6A2 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF521801 // c0 (high single) - .long 0xB2AE4178 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEA72938 // c1 (low single) - .long 0xBFAFCC22 // c2 - .long 0x3FD7BD4A // c3 - .long 0xBFEBB01B // c4 - .long 0xBF687A55 // B' = pi/2 - B (high single) - .long 0x3252257B // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF47C8CC // c0 (high single) - .long 0x3200F51A // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEC82C6C // c1 (low single) - .long 0xBFA0BAE9 // c2 - .long 0x3FC2252F // c3 - .long 0xBFCD24C7 // c4 - .long 0xBF6EC2D4 // B' = pi/2 - B (high single) - .long 0x32BEF0A7 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF3DDCCF // c0 (high single) - .long 0xB2D29606 // c0 (low single) - .long 0x40000000 // c1 (high 1 bit) - .long 0xBEE6606F // c1 (low single) - .long 0xBF9325D6 // c2 - .long 0x3FAF4E69 // c3 - .long 0xBFB3080C // c4 - .long 0xBF750B52 // B' = pi/2 - B (high single) - .long 0xB2EB316F // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF344BA9 // c0 (high single) - .long 0x32B8B0EA // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EFDF4F7 // c1 (low single) - .long 0xBF86DCA8 // c2 - .long 0x3F9ED53B // c3 - .long 0xBF9CBEDE // c4 - .long 0xBF7B53D1 // B' = pi/2 - B (high single) - .long 0xB2955386 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF2B0DC1 // c0 (high single) - .long 0xB2AB7EBA // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EE496C2 // c1 (low single) - .long 0xBF776C40 // c2 - .long 0x3F9065C1 // c3 - .long 0xBF89AFB6 // c4 - .long 0xBF80CE28 // B' = pi/2 - B (high single) - .long 0xB1FDD672 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF221C37 // c0 (high single) - .long 0x320C61DC // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3ECD4F71 // c1 (low single) - .long 0xBF631DAA // c2 - .long 0x3F83B471 // c3 - .long 0xBF7281EA // c4 - .long 0xBF83F267 // B' = pi/2 - B (high single) - .long 0xB374CBD9 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF1970C4 // c0 (high single) - .long 0xB2904848 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EB7EFF8 // c1 (low single) - .long 0xBF50907C // c2 - .long 0x3F710FEA // c3 - .long 0xBF561FED // c4 - .long 0xBF8716A7 // B' = pi/2 - B (high single) - .long 0x32588C6D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF1105AF // c0 (high single) - .long 0xB2F045B0 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3EA44EE2 // c1 (low single) - .long 0xBF3F8FDB // c2 - .long 0x3F5D3FD0 // c3 - .long 0xBF3D0A23 // c4 - .long 0xBF8A3AE6 // B' = pi/2 - B (high single) - .long 0xB31EEDF0 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF08D5B9 // c0 (high single) - .long 0x325EF98E // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E92478D // c1 (low single) - .long 0xBF2FEDC9 // c2 - .long 0x3F4BCD58 // c3 - .long 0xBF27AE9E // c4 - .long 0xBF8D5F26 // B' = pi/2 - B (high single) - .long 0x330C0105 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBF00DC0D // c0 (high single) - .long 0x3214AF72 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E81B994 // c1 (low single) - .long 0xBF218233 // c2 - .long 0x3F3C4531 // c3 - .long 0xBF149688 // c4 - .long 0xBF908365 // B' = pi/2 - B (high single) - .long 0xB292200D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBEF22870 // c0 (high single) - .long 0xB25271F4 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E65107A // c1 (low single) - .long 0xBF1429F0 // c2 - .long 0x3F2E8AFC // c3 - .long 0xBF040498 // c4 - .long 0xBF93A7A5 // B' = pi/2 - B (high single) - .long 0x3361DEEE // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBEE2F439 // c0 (high single) - .long 0x31F4399E // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E49341C // c1 (low single) - .long 0xBF07C61A // c2 - .long 0x3F22560F // c3 - .long 0xBEEAA81E // c4 - .long 0xBF96CBE4 // B' = pi/2 - B (high single) - .long 0x314CDE2E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBED413CD // c0 (high single) - .long 0x31C06152 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E2FB0CC // c1 (low single) - .long 0xBEF876CB // c2 - .long 0x3F177807 // c3 - .long 0xBED08437 // c4 - .long 0xBF99F023 // B' = pi/2 - B (high single) - .long 0xB3484328 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBEC5800D // c0 (high single) - .long 0x3214C3C1 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E185E54 // c1 (low single) - .long 0xBEE2E342 // c2 - .long 0x3F0DCA73 // c3 - .long 0xBEB8CC21 // c4 - .long 0xBF9D1463 // B' = pi/2 - B (high single) - .long 0x32C55799 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBEB73250 // c0 (high single) - .long 0x32028823 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3E0318F8 // c1 (low single) - .long 0xBECEA678 // c2 - .long 0x3F053C67 // c3 - .long 0xBEA41E53 // c4 - .long 0xBFA038A2 // B' = pi/2 - B (high single) - .long 0xB2E4CA7E // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBEA92457 // c0 (high single) - .long 0xB0B80830 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3DDF8200 // c1 (low single) - .long 0xBEBB99E9 // c2 - .long 0x3EFB4AA8 // c3 - .long 0xBE9182BE // c4 - .long 0xBFA35CE2 // B' = pi/2 - B (high single) - .long 0x333889B6 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE9B5042 // c0 (high single) - .long 0x322A3AEE // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3DBC7490 // c1 (low single) - .long 0xBEA99AF5 // c2 - .long 0x3EEDE107 // c3 - .long 0xBE80E9AA // c4 - .long 0xBFA68121 // B' = pi/2 - B (high single) - .long 0xB1E43AAC // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE8DB082 // c0 (high single) - .long 0x3132A234 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D9CD7D0 // c1 (low single) - .long 0xBE988A60 // c2 - .long 0x3EE203E3 // c3 - .long 0xBE63582C // c4 - .long 0xBFA9A560 // B' = pi/2 - B (high single) - .long 0xB3719861 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE803FD4 // c0 (high single) - .long 0x32279E66 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D807FC8 // c1 (low single) - .long 0xBE884BD4 // c2 - .long 0x3ED7812D // c3 - .long 0xBE4636EB // c4 - .long 0xBFACC9A0 // B' = pi/2 - B (high single) - .long 0x32655A50 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE65F267 // c0 (high single) - .long 0xB1B4B1DF // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D4E8B90 // c1 (low single) - .long 0xBE718ACA // c2 - .long 0x3ECE7164 // c3 - .long 0xBE2DC161 // c4 - .long 0xBFAFEDDF // B' = pi/2 - B (high single) - .long 0xB31BBA77 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE4BAFAF // c0 (high single) - .long 0xAF2A29E0 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3D221018 // c1 (low single) - .long 0xBE53BED0 // c2 - .long 0x3EC67E26 // c3 - .long 0xBE1568E2 // c4 - .long 0xBFB3121F // B' = pi/2 - B (high single) - .long 0x330F347D // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE31AE4D // c0 (high single) - .long 0x31F32251 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3CF6A500 // c1 (low single) - .long 0xBE3707DA // c2 - .long 0x3EBFA489 // c3 - .long 0xBDFBD9C7 // c4 - .long 0xBFB6365E // B' = pi/2 - B (high single) - .long 0xB28BB91C // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBE17E564 // c0 (high single) - .long 0x31C5A2E4 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3CB440D0 // c1 (low single) - .long 0xBE1B3D00 // c2 - .long 0x3EB9F664 // c3 - .long 0xBDD647C0 // c4 - .long 0xBFB95A9E // B' = pi/2 - B (high single) - .long 0x33651267 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBDFC98C2 // c0 (high single) - .long 0x30AE525C // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3C793D20 // c1 (low single) - .long 0xBE003845 // c2 - .long 0x3EB5271F // c3 - .long 0xBDAC669E // c4 - .long 0xBFBC7EDD // B' = pi/2 - B (high single) - .long 0x31800ADD // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBDC9B5DC // c0 (high single) - .long 0xB145AD86 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3C1EEF20 // c1 (low single) - .long 0xBDCBAAEA // c2 - .long 0x3EB14E5E // c3 - .long 0xBD858BB2 // c4 - .long 0xBFBFA31C // B' = pi/2 - B (high single) - .long 0xB3450FB0 // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBD9711CE // c0 (high single) - .long 0xB14FEB28 // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3BB24C00 // c1 (low single) - .long 0xBD97E43A // c2 - .long 0x3EAE6A89 // c3 - .long 0xBD4D07E0 // c4 - .long 0xBFC2C75C // B' = pi/2 - B (high single) - .long 0x32CBBE8A // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBD49393C // c0 (high single) - .long 0xB0A39F5B // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3B1E2B00 // c1 (low single) - .long 0xBD49B5D4 // c2 - .long 0x3EAC4F10 // c3 - .long 0xBCFD9425 // c4 - .long 0xBFC5EB9B // B' = pi/2 - B (high single) - .long 0xB2DE638C // B' = pi/2 - B (low single) - .long 0x00000000 // tau (1 for cot path) - .long 0xBCC91A31 // c0 (high single) - .long 0xAF8E8D1A // c0 (low single) - .long 0x3F800000 // c1 (high 1 bit) - .long 0x3A1DFA00 // c1 (low single) - .long 0xBCC9392D // c2 - .long 0x3EAB1889 // c3 - .long 0xBC885D3B // c4 - .align 32 - .type __svml_stan_data_internal, @object - .size __svml_stan_data_internal, .-__svml_stan_data_internal - .align 32 + blsrl %ebx, %ebx + jnz L(SPECIAL_VALUES_LOOP) + + /* All results have been written to (%rsp). */ + vmovups (%rsp), %ymm0 + /* Restore rsp. */ + movq %r13, %rsp + cfi_def_cfa_register (rsp) + /* Restore callee save registers. */ + popq %rbp + cfi_adjust_cfa_offset (-8) + cfi_restore (rbp) + popq %rbx + cfi_adjust_cfa_offset (-8) + cfi_restore (rbp) + popq %r13 + cfi_adjust_cfa_offset (-8) + cfi_restore (r13) + ret +END(_ZGVdN8v_tanf_avx2) -#ifdef __svml_stan_reduction_data_internal_typedef -typedef unsigned int VUINT32; -typedef struct { - __declspec(align(32)) VUINT32 _sPtable[256][3][1]; -} __svml_stan_reduction_data_internal; -#endif -__svml_stan_reduction_data_internal: - /* P_hi P_med P_lo */ - .long 0x00000000, 0x00000000, 0x00000000 /* 0 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 1 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 2 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 3 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 4 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 5 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 6 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 7 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 8 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 9 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 10 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 11 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 12 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 13 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 14 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 15 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 16 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 17 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 18 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 19 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 20 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 21 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 22 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 23 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 24 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 25 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 26 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 27 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 28 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 29 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 30 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 31 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 32 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 33 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 34 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 35 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 36 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 37 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 38 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 39 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 40 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 41 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 42 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 43 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 44 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 45 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 46 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 47 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 48 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 49 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 50 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 51 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 52 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 53 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 54 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 55 */ - .long 0x00000000, 0x00000000, 0x00000000 /* 56 */ - .long 0x00000000, 0x00000000, 0x00000001 /* 57 */ - .long 0x00000000, 0x00000000, 0x00000002 /* 58 */ - .long 0x00000000, 0x00000000, 0x00000005 /* 59 */ - .long 0x00000000, 0x00000000, 0x0000000A /* 60 */ - .long 0x00000000, 0x00000000, 0x00000014 /* 61 */ - .long 0x00000000, 0x00000000, 0x00000028 /* 62 */ - .long 0x00000000, 0x00000000, 0x00000051 /* 63 */ - .long 0x00000000, 0x00000000, 0x000000A2 /* 64 */ - .long 0x00000000, 0x00000000, 0x00000145 /* 65 */ - .long 0x00000000, 0x00000000, 0x0000028B /* 66 */ - .long 0x00000000, 0x00000000, 0x00000517 /* 67 */ - .long 0x00000000, 0x00000000, 0x00000A2F /* 68 */ - .long 0x00000000, 0x00000000, 0x0000145F /* 69 */ - .long 0x00000000, 0x00000000, 0x000028BE /* 70 */ - .long 0x00000000, 0x00000000, 0x0000517C /* 71 */ - .long 0x00000000, 0x00000000, 0x0000A2F9 /* 72 */ - .long 0x00000000, 0x00000000, 0x000145F3 /* 73 */ - .long 0x00000000, 0x00000000, 0x00028BE6 /* 74 */ - .long 0x00000000, 0x00000000, 0x000517CC /* 75 */ - .long 0x00000000, 0x00000000, 0x000A2F98 /* 76 */ - .long 0x00000000, 0x00000000, 0x00145F30 /* 77 */ - .long 0x00000000, 0x00000000, 0x0028BE60 /* 78 */ - .long 0x00000000, 0x00000000, 0x00517CC1 /* 79 */ - .long 0x00000000, 0x00000000, 0x00A2F983 /* 80 */ - .long 0x00000000, 0x00000000, 0x0145F306 /* 81 */ - .long 0x00000000, 0x00000000, 0x028BE60D /* 82 */ - .long 0x00000000, 0x00000000, 0x0517CC1B /* 83 */ - .long 0x00000000, 0x00000000, 0x0A2F9836 /* 84 */ - .long 0x00000000, 0x00000000, 0x145F306D /* 85 */ - .long 0x00000000, 0x00000000, 0x28BE60DB /* 86 */ - .long 0x00000000, 0x00000000, 0x517CC1B7 /* 87 */ - .long 0x00000000, 0x00000000, 0xA2F9836E /* 88 */ - .long 0x00000000, 0x00000001, 0x45F306DC /* 89 */ - .long 0x00000000, 0x00000002, 0x8BE60DB9 /* 90 */ - .long 0x00000000, 0x00000005, 0x17CC1B72 /* 91 */ - .long 0x00000000, 0x0000000A, 0x2F9836E4 /* 92 */ - .long 0x00000000, 0x00000014, 0x5F306DC9 /* 93 */ - .long 0x00000000, 0x00000028, 0xBE60DB93 /* 94 */ - .long 0x00000000, 0x00000051, 0x7CC1B727 /* 95 */ - .long 0x00000000, 0x000000A2, 0xF9836E4E /* 96 */ - .long 0x00000000, 0x00000145, 0xF306DC9C /* 97 */ - .long 0x00000000, 0x0000028B, 0xE60DB939 /* 98 */ - .long 0x00000000, 0x00000517, 0xCC1B7272 /* 99 */ - .long 0x00000000, 0x00000A2F, 0x9836E4E4 /* 100 */ - .long 0x00000000, 0x0000145F, 0x306DC9C8 /* 101 */ - .long 0x00000000, 0x000028BE, 0x60DB9391 /* 102 */ - .long 0x00000000, 0x0000517C, 0xC1B72722 /* 103 */ - .long 0x00000000, 0x0000A2F9, 0x836E4E44 /* 104 */ - .long 0x00000000, 0x000145F3, 0x06DC9C88 /* 105 */ - .long 0x00000000, 0x00028BE6, 0x0DB93910 /* 106 */ - .long 0x00000000, 0x000517CC, 0x1B727220 /* 107 */ - .long 0x00000000, 0x000A2F98, 0x36E4E441 /* 108 */ - .long 0x00000000, 0x00145F30, 0x6DC9C882 /* 109 */ - .long 0x00000000, 0x0028BE60, 0xDB939105 /* 110 */ - .long 0x00000000, 0x00517CC1, 0xB727220A /* 111 */ - .long 0x00000000, 0x00A2F983, 0x6E4E4415 /* 112 */ - .long 0x00000000, 0x0145F306, 0xDC9C882A /* 113 */ - .long 0x00000000, 0x028BE60D, 0xB9391054 /* 114 */ - .long 0x00000000, 0x0517CC1B, 0x727220A9 /* 115 */ - .long 0x00000000, 0x0A2F9836, 0xE4E44152 /* 116 */ - .long 0x00000000, 0x145F306D, 0xC9C882A5 /* 117 */ - .long 0x00000000, 0x28BE60DB, 0x9391054A /* 118 */ - .long 0x00000000, 0x517CC1B7, 0x27220A94 /* 119 */ - .long 0x00000000, 0xA2F9836E, 0x4E441529 /* 120 */ - .long 0x00000001, 0x45F306DC, 0x9C882A53 /* 121 */ - .long 0x00000002, 0x8BE60DB9, 0x391054A7 /* 122 */ - .long 0x00000005, 0x17CC1B72, 0x7220A94F /* 123 */ - .long 0x0000000A, 0x2F9836E4, 0xE441529F /* 124 */ - .long 0x00000014, 0x5F306DC9, 0xC882A53F /* 125 */ - .long 0x00000028, 0xBE60DB93, 0x91054A7F /* 126 */ - .long 0x00000051, 0x7CC1B727, 0x220A94FE /* 127 */ - .long 0x000000A2, 0xF9836E4E, 0x441529FC /* 128 */ - .long 0x00000145, 0xF306DC9C, 0x882A53F8 /* 129 */ - .long 0x0000028B, 0xE60DB939, 0x1054A7F0 /* 130 */ - .long 0x00000517, 0xCC1B7272, 0x20A94FE1 /* 131 */ - .long 0x00000A2F, 0x9836E4E4, 0x41529FC2 /* 132 */ - .long 0x0000145F, 0x306DC9C8, 0x82A53F84 /* 133 */ - .long 0x000028BE, 0x60DB9391, 0x054A7F09 /* 134 */ - .long 0x0000517C, 0xC1B72722, 0x0A94FE13 /* 135 */ - .long 0x0000A2F9, 0x836E4E44, 0x1529FC27 /* 136 */ - .long 0x000145F3, 0x06DC9C88, 0x2A53F84E /* 137 */ - .long 0x00028BE6, 0x0DB93910, 0x54A7F09D /* 138 */ - .long 0x000517CC, 0x1B727220, 0xA94FE13A /* 139 */ - .long 0x000A2F98, 0x36E4E441, 0x529FC275 /* 140 */ - .long 0x00145F30, 0x6DC9C882, 0xA53F84EA /* 141 */ - .long 0x0028BE60, 0xDB939105, 0x4A7F09D5 /* 142 */ - .long 0x00517CC1, 0xB727220A, 0x94FE13AB /* 143 */ - .long 0x00A2F983, 0x6E4E4415, 0x29FC2757 /* 144 */ - .long 0x0145F306, 0xDC9C882A, 0x53F84EAF /* 145 */ - .long 0x028BE60D, 0xB9391054, 0xA7F09D5F /* 146 */ - .long 0x0517CC1B, 0x727220A9, 0x4FE13ABE /* 147 */ - .long 0x0A2F9836, 0xE4E44152, 0x9FC2757D /* 148 */ - .long 0x145F306D, 0xC9C882A5, 0x3F84EAFA /* 149 */ - .long 0x28BE60DB, 0x9391054A, 0x7F09D5F4 /* 150 */ - .long 0x517CC1B7, 0x27220A94, 0xFE13ABE8 /* 151 */ - .long 0xA2F9836E, 0x4E441529, 0xFC2757D1 /* 152 */ - .long 0x45F306DC, 0x9C882A53, 0xF84EAFA3 /* 153 */ - .long 0x8BE60DB9, 0x391054A7, 0xF09D5F47 /* 154 */ - .long 0x17CC1B72, 0x7220A94F, 0xE13ABE8F /* 155 */ - .long 0x2F9836E4, 0xE441529F, 0xC2757D1F /* 156 */ - .long 0x5F306DC9, 0xC882A53F, 0x84EAFA3E /* 157 */ - .long 0xBE60DB93, 0x91054A7F, 0x09D5F47D /* 158 */ - .long 0x7CC1B727, 0x220A94FE, 0x13ABE8FA /* 159 */ - .long 0xF9836E4E, 0x441529FC, 0x2757D1F5 /* 160 */ - .long 0xF306DC9C, 0x882A53F8, 0x4EAFA3EA /* 161 */ - .long 0xE60DB939, 0x1054A7F0, 0x9D5F47D4 /* 162 */ - .long 0xCC1B7272, 0x20A94FE1, 0x3ABE8FA9 /* 163 */ - .long 0x9836E4E4, 0x41529FC2, 0x757D1F53 /* 164 */ - .long 0x306DC9C8, 0x82A53F84, 0xEAFA3EA6 /* 165 */ - .long 0x60DB9391, 0x054A7F09, 0xD5F47D4D /* 166 */ - .long 0xC1B72722, 0x0A94FE13, 0xABE8FA9A /* 167 */ - .long 0x836E4E44, 0x1529FC27, 0x57D1F534 /* 168 */ - .long 0x06DC9C88, 0x2A53F84E, 0xAFA3EA69 /* 169 */ - .long 0x0DB93910, 0x54A7F09D, 0x5F47D4D3 /* 170 */ - .long 0x1B727220, 0xA94FE13A, 0xBE8FA9A6 /* 171 */ - .long 0x36E4E441, 0x529FC275, 0x7D1F534D /* 172 */ - .long 0x6DC9C882, 0xA53F84EA, 0xFA3EA69B /* 173 */ - .long 0xDB939105, 0x4A7F09D5, 0xF47D4D37 /* 174 */ - .long 0xB727220A, 0x94FE13AB, 0xE8FA9A6E /* 175 */ - .long 0x6E4E4415, 0x29FC2757, 0xD1F534DD /* 176 */ - .long 0xDC9C882A, 0x53F84EAF, 0xA3EA69BB /* 177 */ - .long 0xB9391054, 0xA7F09D5F, 0x47D4D377 /* 178 */ - .long 0x727220A9, 0x4FE13ABE, 0x8FA9A6EE /* 179 */ - .long 0xE4E44152, 0x9FC2757D, 0x1F534DDC /* 180 */ - .long 0xC9C882A5, 0x3F84EAFA, 0x3EA69BB8 /* 181 */ - .long 0x9391054A, 0x7F09D5F4, 0x7D4D3770 /* 182 */ - .long 0x27220A94, 0xFE13ABE8, 0xFA9A6EE0 /* 183 */ - .long 0x4E441529, 0xFC2757D1, 0xF534DDC0 /* 184 */ - .long 0x9C882A53, 0xF84EAFA3, 0xEA69BB81 /* 185 */ - .long 0x391054A7, 0xF09D5F47, 0xD4D37703 /* 186 */ - .long 0x7220A94F, 0xE13ABE8F, 0xA9A6EE06 /* 187 */ - .long 0xE441529F, 0xC2757D1F, 0x534DDC0D /* 188 */ - .long 0xC882A53F, 0x84EAFA3E, 0xA69BB81B /* 189 */ - .long 0x91054A7F, 0x09D5F47D, 0x4D377036 /* 190 */ - .long 0x220A94FE, 0x13ABE8FA, 0x9A6EE06D /* 191 */ - .long 0x441529FC, 0x2757D1F5, 0x34DDC0DB /* 192 */ - .long 0x882A53F8, 0x4EAFA3EA, 0x69BB81B6 /* 193 */ - .long 0x1054A7F0, 0x9D5F47D4, 0xD377036D /* 194 */ - .long 0x20A94FE1, 0x3ABE8FA9, 0xA6EE06DB /* 195 */ - .long 0x41529FC2, 0x757D1F53, 0x4DDC0DB6 /* 196 */ - .long 0x82A53F84, 0xEAFA3EA6, 0x9BB81B6C /* 197 */ - .long 0x054A7F09, 0xD5F47D4D, 0x377036D8 /* 198 */ - .long 0x0A94FE13, 0xABE8FA9A, 0x6EE06DB1 /* 199 */ - .long 0x1529FC27, 0x57D1F534, 0xDDC0DB62 /* 200 */ - .long 0x2A53F84E, 0xAFA3EA69, 0xBB81B6C5 /* 201 */ - .long 0x54A7F09D, 0x5F47D4D3, 0x77036D8A /* 202 */ - .long 0xA94FE13A, 0xBE8FA9A6, 0xEE06DB14 /* 203 */ - .long 0x529FC275, 0x7D1F534D, 0xDC0DB629 /* 204 */ - .long 0xA53F84EA, 0xFA3EA69B, 0xB81B6C52 /* 205 */ - .long 0x4A7F09D5, 0xF47D4D37, 0x7036D8A5 /* 206 */ - .long 0x94FE13AB, 0xE8FA9A6E, 0xE06DB14A /* 207 */ - .long 0x29FC2757, 0xD1F534DD, 0xC0DB6295 /* 208 */ - .long 0x53F84EAF, 0xA3EA69BB, 0x81B6C52B /* 209 */ - .long 0xA7F09D5F, 0x47D4D377, 0x036D8A56 /* 210 */ - .long 0x4FE13ABE, 0x8FA9A6EE, 0x06DB14AC /* 211 */ - .long 0x9FC2757D, 0x1F534DDC, 0x0DB62959 /* 212 */ - .long 0x3F84EAFA, 0x3EA69BB8, 0x1B6C52B3 /* 213 */ - .long 0x7F09D5F4, 0x7D4D3770, 0x36D8A566 /* 214 */ - .long 0xFE13ABE8, 0xFA9A6EE0, 0x6DB14ACC /* 215 */ - .long 0xFC2757D1, 0xF534DDC0, 0xDB629599 /* 216 */ - .long 0xF84EAFA3, 0xEA69BB81, 0xB6C52B32 /* 217 */ - .long 0xF09D5F47, 0xD4D37703, 0x6D8A5664 /* 218 */ - .long 0xE13ABE8F, 0xA9A6EE06, 0xDB14ACC9 /* 219 */ - .long 0xC2757D1F, 0x534DDC0D, 0xB6295993 /* 220 */ - .long 0x84EAFA3E, 0xA69BB81B, 0x6C52B327 /* 221 */ - .long 0x09D5F47D, 0x4D377036, 0xD8A5664F /* 222 */ - .long 0x13ABE8FA, 0x9A6EE06D, 0xB14ACC9E /* 223 */ - .long 0x2757D1F5, 0x34DDC0DB, 0x6295993C /* 224 */ - .long 0x4EAFA3EA, 0x69BB81B6, 0xC52B3278 /* 225 */ - .long 0x9D5F47D4, 0xD377036D, 0x8A5664F1 /* 226 */ - .long 0x3ABE8FA9, 0xA6EE06DB, 0x14ACC9E2 /* 227 */ - .long 0x757D1F53, 0x4DDC0DB6, 0x295993C4 /* 228 */ - .long 0xEAFA3EA6, 0x9BB81B6C, 0x52B32788 /* 229 */ - .long 0xD5F47D4D, 0x377036D8, 0xA5664F10 /* 230 */ - .long 0xABE8FA9A, 0x6EE06DB1, 0x4ACC9E21 /* 231 */ - .long 0x57D1F534, 0xDDC0DB62, 0x95993C43 /* 232 */ - .long 0xAFA3EA69, 0xBB81B6C5, 0x2B327887 /* 233 */ - .long 0x5F47D4D3, 0x77036D8A, 0x5664F10E /* 234 */ - .long 0xBE8FA9A6, 0xEE06DB14, 0xACC9E21C /* 235 */ - .long 0x7D1F534D, 0xDC0DB629, 0x5993C439 /* 236 */ - .long 0xFA3EA69B, 0xB81B6C52, 0xB3278872 /* 237 */ - .long 0xF47D4D37, 0x7036D8A5, 0x664F10E4 /* 238 */ - .long 0xE8FA9A6E, 0xE06DB14A, 0xCC9E21C8 /* 239 */ - .long 0xD1F534DD, 0xC0DB6295, 0x993C4390 /* 240 */ - .long 0xA3EA69BB, 0x81B6C52B, 0x32788720 /* 241 */ - .long 0x47D4D377, 0x036D8A56, 0x64F10E41 /* 242 */ - .long 0x8FA9A6EE, 0x06DB14AC, 0xC9E21C82 /* 243 */ - .long 0x1F534DDC, 0x0DB62959, 0x93C43904 /* 244 */ - .long 0x3EA69BB8, 0x1B6C52B3, 0x27887208 /* 245 */ - .long 0x7D4D3770, 0x36D8A566, 0x4F10E410 /* 246 */ - .long 0xFA9A6EE0, 0x6DB14ACC, 0x9E21C820 /* 247 */ - .long 0xF534DDC0, 0xDB629599, 0x3C439041 /* 248 */ - .long 0xEA69BB81, 0xB6C52B32, 0x78872083 /* 249 */ - .long 0xD4D37703, 0x6D8A5664, 0xF10E4107 /* 250 */ - .long 0xA9A6EE06, 0xDB14ACC9, 0xE21C820F /* 251 */ - .long 0x534DDC0D, 0xB6295993, 0xC439041F /* 252 */ - .long 0xA69BB81B, 0x6C52B327, 0x8872083F /* 253 */ - .long 0x4D377036, 0xD8A5664F, 0x10E4107F /* 254 */ - .long 0x9A6EE06D, 0xB14ACC9E, 0x21C820FF /* 255 */ + .section .rodata.avx2, "a" .align 32 - .type __svml_stan_reduction_data_internal, @object - .size __svml_stan_reduction_data_internal, .-__svml_stan_reduction_data_internal +LOCAL_DATA_NAME: + + DATA_VEC (LOCAL_DATA_NAME, _sPI2_FMA, 0xB33BBD2E) // AVX2 + DATA_VEC (LOCAL_DATA_NAME, _sPI3_FMA, 0xA6F72CED) // AVX2 + DATA_VEC (LOCAL_DATA_NAME, _FLT_0, 0x7f800000) // AVX512, AVX2 + DATA_VEC (LOCAL_DATA_NAME, _FLT_1, 0x00000080) // AVX2 + DATA_VEC (LOCAL_DATA_NAME, _FLT_2, 0x35800000) // AVX512, AVX2 + DATA_VEC (LOCAL_DATA_NAME, _FLT_3, 0xb43bbd2e) // AVX512, AVX2 + + .type LOCAL_DATA_NAME, @object + .size LOCAL_DATA_NAME, .-LOCAL_DATA_NAME

[v1,17/27] x86/fpu: Optimize svml_s_tanf8_core_avx2.S

Commit Message

Patch