[v1,16/27] x86/fpu: Optimize svml_s_tanf4_core_sse4.S

Message ID 20221207085236.1424424-16-goldstein.w.n@gmail.com
State New
Headers
Series [v1,01/27] x86/fpu: Create helper file for common data macros |

Checks

Context Check Description
dj/TryBot-apply_patch success Patch applied to master at the time it was sent
redhat-pt-bot/TryBot-still_applies warning Patch no longer applies to master

Commit Message

Noah Goldstein Dec. 7, 2022, 8:52 a.m. UTC
  1. Remove many unnecissary spills.
2. Cleanup some missed optimizations in instruction selection /
   unnecissary repeated rodata references.
3. Remove unused rodata.
4. Use common data definitions where possible.

Code Size Change: -980 Bytes (1619 - 2599)

Input                                 New Time / Old Time
0F          (0x00000000)           -> 0.8527
0F          (0x0000ffff, Denorm)   -> 0.9879
.1F         (0x3dcccccd)           -> 0.8542
5F          (0x40a00000)           -> 0.8633
2315255808F (0x4f0a0000)           -> 0.7640
-NaN        (0xffffffff)           -> 0.7966
---
 .../fpu/multiarch/svml_s_tanf4_core_sse4.S    | 3031 +++--------------
 1 file changed, 531 insertions(+), 2500 deletions(-)
  

Patch

diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_tanf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_tanf4_core_sse4.S
index 3dc82cae68..f3f0c867ef 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_tanf4_core_sse4.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_tanf4_core_sse4.S
@@ -45,2553 +45,584 @@ 
  *
  */
 
-/* Offsets for data table __svml_stan_data_internal
- */
-#define _sInvPI_uisa			0
-#define _sPI1_uisa			16
-#define _sPI2_uisa			32
-#define _sPI3_uisa			48
-#define _sPI2_ha_uisa			64
-#define _sPI3_ha_uisa			80
-#define Th_tbl_uisa			96
-#define Tl_tbl_uisa			224
-#define _sPC3_uisa			352
-#define _sPC5_uisa			368
-#define _sRangeReductionVal_uisa	384
-#define _sInvPi				400
-#define _sSignMask			416
-#define _sAbsMask			432
-#define _sRangeVal			448
-#define _sRShifter			464
-#define _sOne				480
-#define _sRangeReductionVal		496
-#define _sPI1				512
-#define _sPI2				528
-#define _sPI3				544
-#define _sPI4				560
-#define _sPI1_FMA			576
-#define _sPI2_FMA			592
-#define _sPI3_FMA			608
-#define _sP0				624
-#define _sP1				640
-#define _sQ0				656
-#define _sQ1				672
-#define _sQ2				688
-#define _sTwo				704
-#define _sCoeffs			720
+#define LOCAL_DATA_NAME	__svml_stan_data_internal
+#include "svml_s_common_sse4_rodata_offsets.h"
+
+#define AVX2_SHARED_OFFSETS
+#define AVX512_SHARED_OFFSETS
+#include "svml_s_tanf_rodata.h.S"
+
+/* Offsets for data table __svml_stan_data_internal.  */
+#define _sPI1	0
+#define _sPI2	16
+#define _sPI3	32
+#define _sPI4	48
+#define _sRangeVal	64
+#define _FLT_0	80
+#define _FLT_1	96
+
 
 #include <sysdep.h>
 
 	.section .text.sse4, "ax", @progbits
 ENTRY(_ZGVbN4v_tanf_sse4)
-	subq	$232, %rsp
-	cfi_def_cfa_offset(240)
-	movaps	%xmm0, %xmm13
-	movups	_sAbsMask+__svml_stan_data_internal(%rip), %xmm12
-
-	/*
-	 * Legacy Code
-	 * Here HW FMA can be unavailable
-	 */
-	xorl	%eax, %eax
-	movaps	%xmm12, %xmm4
-	pxor	%xmm10, %xmm10
-	movups	_sInvPi+__svml_stan_data_internal(%rip), %xmm2
-	andps	%xmm13, %xmm4
-	mulps	%xmm4, %xmm2
-
-	/* Range reduction */
-	movaps	%xmm4, %xmm1
-
-	/*
-	 *
-	 * Main path (_LA_ and _EP_)
-	 *
-	 * Octant calculation
-	 */
-	movups	_sRShifter+__svml_stan_data_internal(%rip), %xmm3
-
-	/* Large values check */
-	movaps	%xmm4, %xmm11
-	movups	_sPI1+__svml_stan_data_internal(%rip), %xmm5
-	andnps	%xmm13, %xmm12
-	movups	_sPI2+__svml_stan_data_internal(%rip), %xmm6
-	addps	%xmm3, %xmm2
-	cmpnleps _sRangeReductionVal+__svml_stan_data_internal(%rip), %xmm11
-	movaps	%xmm2, %xmm8
-	movups	_sPI3+__svml_stan_data_internal(%rip), %xmm7
-	subps	%xmm3, %xmm8
-	movmskps %xmm11, %edx
-	movups	_sPI4+__svml_stan_data_internal(%rip), %xmm9
-	mulps	%xmm8, %xmm5
-	mulps	%xmm8, %xmm6
-	mulps	%xmm8, %xmm7
-	subps	%xmm5, %xmm1
-	mulps	%xmm8, %xmm9
-	subps	%xmm6, %xmm1
-	movups	_sQ2+__svml_stan_data_internal(%rip), %xmm15
+	movaps	%xmm0, %xmm15
+	movups	COMMON_DATA(_AbsMask)(%rip), %xmm4
 
-	/* Inversion mask and sign calculation */
-	movaps	%xmm2, %xmm5
+	andps	%xmm0, %xmm4
 
-	/* Rational approximation */
-	movups	_sP1+__svml_stan_data_internal(%rip), %xmm14
-	pslld	$30, %xmm2
-	cmpneqps %xmm10, %xmm2
-	subps	%xmm7, %xmm1
+	movups	AVX2_SHARED_DATA(_sInvPi)(%rip), %xmm0
+	mulps	%xmm4, %xmm0
 
-	/* Exchanged numerator and denominator if necessary */
-	movaps	%xmm2, %xmm0
-	movaps	%xmm2, %xmm10
-	pslld	$31, %xmm5
-	subps	%xmm9, %xmm1
-	movaps	%xmm1, %xmm3
-	pxor	%xmm12, %xmm5
-	mulps	%xmm1, %xmm3
-	mulps	%xmm3, %xmm15
-	mulps	%xmm3, %xmm14
-	addps	_sQ1+__svml_stan_data_internal(%rip), %xmm15
-	addps	_sP0+__svml_stan_data_internal(%rip), %xmm14
-	mulps	%xmm15, %xmm3
-	mulps	%xmm14, %xmm1
-	addps	_sQ0+__svml_stan_data_internal(%rip), %xmm3
-	andnps	%xmm1, %xmm0
-	andps	%xmm3, %xmm10
-	andps	%xmm2, %xmm1
-	andnps	%xmm3, %xmm2
-	orps	%xmm10, %xmm0
-	orps	%xmm2, %xmm1
-
-	/* Division */
-	divps	%xmm1, %xmm0
-
-	/* Sign setting */
-	pxor	%xmm5, %xmm0
+	/* Range reduction.  */
+	movaps	%xmm4, %xmm1
 
 	/*
-	 *
-	 * End of main path (_LA_ and _EP_)
-	 */
+	   Main path (_LA_ and _EP_)
 
-	testl	%edx, %edx
+	   Octant calculation.  */
+	movups	AVX2_SHARED_DATA(_sRShifter)(%rip), %xmm3
 
-	/* Go to auxilary branch */
-	jne	L(AUX_BRANCH)
-	# LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm4 xmm11 xmm12 xmm13
-
-	/* Return from auxilary branch
-	 * for out of main path inputs
-	 */
+	/* Large values check.  */
+	movups	LOCAL_DATA(_sPI1)(%rip), %xmm5
+	movups	LOCAL_DATA(_sPI2)(%rip), %xmm6
+	addps	%xmm3, %xmm0
+	movaps	%xmm0, %xmm2
+	movups	LOCAL_DATA(_sPI3)(%rip), %xmm7
+	subps	%xmm3, %xmm2
 
-L(AUX_BRANCH_RETURN):
-	testl	%eax, %eax
+	mulps	%xmm2, %xmm5
+	mulps	%xmm2, %xmm6
+	mulps	%xmm2, %xmm7
 
-	/* Go to special inputs processing branch */
-	jne	L(SPECIAL_VALUES_BRANCH)
-	# LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm13
+	subps	%xmm5, %xmm1
+	mulps	LOCAL_DATA(_sPI4)(%rip), %xmm2
+	subps	%xmm6, %xmm1
+	movups	AVX2_SHARED_DATA(_sQ2)(%rip), %xmm6
 
-	/* Restore registers
-	 * and exit the function
-	 */
 
-L(EXIT):
-	addq	$232, %rsp
-	cfi_def_cfa_offset(8)
-	ret
-	cfi_def_cfa_offset(240)
+	/* Rational approximation.  */
+	movups	AVX2_SHARED_DATA(_sP1)(%rip), %xmm5
 
-	/* Branch to process
-	 * special inputs
-	 */
+	/* Inversion mask and sign calculation.  */
+	pslld	$31, %xmm0
+	subps	%xmm7, %xmm1
 
-L(SPECIAL_VALUES_BRANCH):
-	movups	%xmm13, 32(%rsp)
-	movups	%xmm0, 48(%rsp)
-	# LOE rbx rbp r12 r13 r14 r15 eax xmm0
-
-	xorl	%edx, %edx
-	movq	%r12, 16(%rsp)
-	cfi_offset(12, -224)
-	movl	%edx, %r12d
-	movq	%r13, 8(%rsp)
-	cfi_offset(13, -232)
-	movl	%eax, %r13d
-	movq	%r14, (%rsp)
-	cfi_offset(14, -240)
-	# LOE rbx rbp r15 r12d r13d
-
-	/* Range mask
-	 * bits check
-	 */
-
-L(RANGEMASK_CHECK):
-	btl	%r12d, %r13d
-
-	/* Call scalar math function */
-	jc	L(SCALAR_MATH_CALL)
-	# LOE rbx rbp r15 r12d r13d
-
-	/* Special inputs
-	 * processing loop
-	 */
+	/* Exchanged numerator and denominator if necessary.  */
+	subps	%xmm2, %xmm1
+	movaps	%xmm1, %xmm3
+	mulps	%xmm1, %xmm1
+	mulps	%xmm1, %xmm6
+	mulps	%xmm1, %xmm5
+	addps	AVX2_SHARED_DATA(_sQ1)(%rip), %xmm6
+	movups	AVX2_SHARED_DATA(_sP0)(%rip), %xmm2
+	addps	%xmm2, %xmm5
+	mulps	%xmm6, %xmm1
+	mulps	%xmm5, %xmm3
+	addps	%xmm2, %xmm1
 
-L(SPECIAL_VALUES_LOOP):
-	incl	%r12d
-	cmpl	$4, %r12d
-
-	/* Check bits in range mask */
-	jl	L(RANGEMASK_CHECK)
-	# LOE rbx rbp r15 r12d r13d
-
-	movq	16(%rsp), %r12
-	cfi_restore(12)
-	movq	8(%rsp), %r13
-	cfi_restore(13)
-	movq	(%rsp), %r14
-	cfi_restore(14)
-	movups	48(%rsp), %xmm0
-
-	/* Go to exit */
-	jmp	L(EXIT)
-	cfi_offset(12, -224)
-	cfi_offset(13, -232)
-	cfi_offset(14, -240)
-	# LOE rbx rbp r12 r13 r14 r15 xmm0
-
-	/* Scalar math fucntion call
-	 * to process special input
-	 */
-
-L(SCALAR_MATH_CALL):
-	movl	%r12d, %r14d
-	movss	32(%rsp, %r14, 4), %xmm0
-	call	tanf@PLT
-	# LOE rbx rbp r14 r15 r12d r13d xmm0
+	movaps	%xmm3, %xmm2
+	blendvps %xmm0, %xmm1, %xmm3
+	blendvps %xmm0, %xmm2, %xmm1
 
-	movss	%xmm0, 48(%rsp, %r14, 4)
+	/* Division.  */
+	divps	%xmm1, %xmm3
 
-	/* Process special inputs in loop */
-	jmp	L(SPECIAL_VALUES_LOOP)
-	cfi_restore(12)
-	cfi_restore(13)
-	cfi_restore(14)
-	# LOE rbx rbp r15 r12d r13d
+	/* Sign setting.  */
+	pxor	%xmm3, %xmm0
 
-	/* Auxilary branch
-	 * for out of main path inputs
-	 */
+	movaps	%xmm4, %xmm3
+	pcmpgtd	AVX2_SHARED_DATA(_sRangeReductionVal)(%rip), %xmm3
+	pmovmskb %xmm3, %edx
 
-L(AUX_BRANCH):
-	movl	$2139095040, %eax
+	/* End of main path (_LA_ and _EP_).  */
+	testl	%edx, %edx
+	/* Go to auxilary branch.  */
+	jne	L(AUX_BRANCH)
 
-	/*
-	 * 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), %r8
-	movups	%xmm13, 64(%rsp)
+	/* Set sign.  */
+	andnps	%xmm15, %xmm4
+	pxor	%xmm4, %xmm0
+	ret
 
-	/*
-	 * 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
-	 */
-	movl	$8388607, %r9d
-	movups	%xmm12, 80(%rsp)
-	movl	$8388608, %r10d
-	movups	%xmm11, 96(%rsp)
+L(AUX_BRANCH):
+	movaps	%xmm3, %xmm14
+	andnps	%xmm0, %xmm3
 
-	/*
-	 * Break the P_xxx and m into 16-bit chunks ready for
-	 * the long multiplication via 16x16->32 multiplications
-	 */
-	movl	$65535, %r11d
-	movd	%eax, %xmm3
-	pshufd	$0, %xmm3, %xmm2
-	andps	%xmm2, %xmm13
-	cmpeqps	%xmm2, %xmm13
-	pand	%xmm4, %xmm2
-	psrld	$23, %xmm2
-	movdqa	%xmm2, %xmm12
-	pslld	$1, %xmm12
-	paddd	%xmm2, %xmm12
-	pslld	$2, %xmm12
-	pshufd	$1, %xmm12, %xmm10
-	pshufd	$2, %xmm12, %xmm11
-	pshufd	$3, %xmm12, %xmm14
-	movd	%xmm12, %edx
-	movd	%xmm10, %ecx
-	movd	%xmm11, %esi
-	movd	%r9d, %xmm11
-	movd	%xmm14, %edi
-	movd	4(%rdx, %r8), %xmm6
-	movd	4(%rcx, %r8), %xmm7
-	movd	4(%rsi, %r8), %xmm3
-	movl	$872415232, %r9d
-	movd	4(%rdi, %r8), %xmm5
-	punpckldq %xmm7, %xmm6
-	punpckldq %xmm5, %xmm3
-	movd	8(%rdi, %r8), %xmm10
-	movmskps %xmm13, %eax
-	punpcklqdq %xmm3, %xmm6
-	movd	8(%rdx, %r8), %xmm3
-	movd	8(%rcx, %r8), %xmm2
-	movd	8(%rsi, %r8), %xmm13
-	punpckldq %xmm2, %xmm3
-	punpckldq %xmm10, %xmm13
-	punpcklqdq %xmm13, %xmm3
-	pshufd	$0, %xmm11, %xmm13
-	movdqa	%xmm3, %xmm2
-	movups	%xmm4, 48(%rsp)
-	pand	%xmm4, %xmm13
-	movd	%r10d, %xmm4
-	psrld	$16, %xmm2
-	movd	(%rdx, %r8), %xmm9
+	/* Get the (2^a / 2pi) mod 1 values from the table.  */
+	movaps	%xmm4, %xmm1
+	psrld	$0x17, %xmm4
+	/* Compute indices in xmm5 (need 4x scale).  */
+	movaps	%xmm4, %xmm5
+	paddd	%xmm4, %xmm4
+	paddd	%xmm4, %xmm5
+
+	pextrq	$0x1, %xmm5, %rcx
+	movq	%xmm5, %rdx
+
+
+	/* Move indices into GPRs.  */
+	movl	%edx, %esi
+	movl	%ecx, %edi
+	shrq	$0x20, %rdx
+	shrq	$0x20, %rcx
+
+	lea	AVX512_SHARED_DATA(_Reduction)(%rip), %rax
+	movq	0(%rax, %rcx, 4), %xmm4
+	movq	0(%rax, %rdi, 4), %xmm5
+	punpckldq %xmm4, %xmm5
+	movq	0(%rax, %rsi, 4), %xmm4
+	movq	0(%rax, %rdx, 4), %xmm2
+	movaps	AVX2_SHARED_DATA(_Low16)(%rip), %xmm9
+	punpckldq %xmm2, %xmm4
+	movaps	%xmm4, %xmm2
+	punpcklqdq %xmm5, %xmm4
+	punpckhqdq %xmm5, %xmm2
+
+	/* Break the P_xxx and m into 16-bit chunks ready for
+	   the long multiplication via 16x16->32 multiplications.  */
+	movaps	%xmm4, %xmm5
+	pand	%xmm9, %xmm4
+	psrld	$0x10, %xmm5
+	movaps	%xmm4, %xmm6
+	psrlq	$0x20, %xmm4
+	movaps	COMMON_DATA(_NotiOffExpoMask)(%rip), %xmm8
+	pandn	%xmm1, %xmm8
+	/* 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.  */
+	movaps	LOCAL_DATA(_sRangeVal)(%rip), %xmm7
+	paddd	%xmm7, %xmm1
+	movmskps %xmm1, %r8d
 
-	/*
-	 * 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).
-	 */
-	movl	$-2147483648, %edx
-	movd	(%rcx, %r8), %xmm8
+	por	%xmm8, %xmm7
 
-	/*
-	 * Create floating-point high part, implicitly adding integer bit 1
-	 * Incorporate overall sign at this stage too.
-	 */
-	movl	$1065353216, %ecx
-	movd	(%rsi, %r8), %xmm15
+	pand	%xmm9, %xmm8
+	movaps	%xmm8, %xmm1
 
-	/*
-	 * 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.
-	 */
-	movl	$1195376640, %esi
-	movd	(%rdi, %r8), %xmm1
-	movl	$511, %r10d
-	movups	%xmm0, 112(%rsp)
-	movd	%r11d, %xmm0
-	pshufd	$0, %xmm4, %xmm12
-	movdqa	%xmm2, %xmm4
-	punpckldq %xmm8, %xmm9
-	paddd	%xmm12, %xmm13
-	punpckldq %xmm1, %xmm15
-	movdqa	%xmm13, %xmm12
-	pshufd	$0, %xmm0, %xmm8
-	movdqa	%xmm6, %xmm0
-	punpcklqdq %xmm15, %xmm9
-	pand	%xmm8, %xmm13
-	movdqa	%xmm9, %xmm14
-	pand	%xmm8, %xmm9
-	movdqa	%xmm13, %xmm10
-	psrld	$16, %xmm14
-	movdqu	%xmm14, 128(%rsp)
-
-	/* Now do the big multiplication and carry propagation */
-	movdqa	%xmm9, %xmm14
-	psrlq	$32, %xmm10
-	psrlq	$32, %xmm14
-	movdqa	%xmm13, %xmm15
-	movdqa	%xmm10, %xmm7
-	pmuludq	%xmm9, %xmm15
-	psrld	$16, %xmm0
-	pmuludq	%xmm14, %xmm7
-	movdqu	%xmm9, 144(%rsp)
-	psllq	$32, %xmm7
-	movdqu	.FLT_16(%rip), %xmm9
-	pand	%xmm8, %xmm6
-	pand	%xmm9, %xmm15
-	psrld	$16, %xmm12
-	movdqa	%xmm0, %xmm1
-	por	%xmm7, %xmm15
-	movdqa	%xmm13, %xmm7
-	pand	%xmm8, %xmm3
-	movdqu	%xmm0, 160(%rsp)
-	movdqa	%xmm12, %xmm11
-	movdqu	%xmm15, 208(%rsp)
-	psrlq	$32, %xmm1
-	pmuludq	%xmm0, %xmm7
-	movdqa	%xmm6, %xmm5
-	movdqa	%xmm10, %xmm15
-	movdqa	%xmm12, %xmm0
-	movdqu	%xmm14, 176(%rsp)
-	psrlq	$32, %xmm11
-	movdqu	%xmm1, 192(%rsp)
-	psrlq	$32, %xmm5
-	pmuludq	%xmm1, %xmm15
-	movdqa	%xmm13, %xmm1
-	pmuludq	%xmm3, %xmm0
+	psrlq	$0x20, %xmm8
+	movaps	%xmm8, %xmm10
+	pmuludq	%xmm4, %xmm8
+	psllq	$0x20, %xmm8
+	movaps	%xmm1, %xmm11
 	pmuludq	%xmm6, %xmm1
-	pmuludq	%xmm12, %xmm6
-	movdqa	%xmm10, %xmm14
-	psrlq	$32, %xmm3
-	pmuludq	%xmm5, %xmm14
-	pand	%xmm9, %xmm1
-	pmuludq	%xmm11, %xmm3
-	pmuludq	%xmm11, %xmm5
-	psllq	$32, %xmm14
-	pand	%xmm9, %xmm0
-	psllq	$32, %xmm3
-	psrlq	$32, %xmm4
-	por	%xmm14, %xmm1
-	por	%xmm3, %xmm0
-	movdqa	%xmm12, %xmm14
-	movdqa	%xmm11, %xmm3
-	pmuludq	%xmm2, %xmm14
-	pand	%xmm9, %xmm7
-	pmuludq	%xmm4, %xmm3
-	pmuludq	%xmm13, %xmm2
-	pmuludq	%xmm10, %xmm4
-	pand	%xmm9, %xmm2
-	psllq	$32, %xmm4
-	psllq	$32, %xmm15
-	pand	%xmm9, %xmm14
-	psllq	$32, %xmm3
-	por	%xmm4, %xmm2
-	por	%xmm15, %xmm7
-	por	%xmm3, %xmm14
-	psrld	$16, %xmm2
-	pand	%xmm9, %xmm6
-	psllq	$32, %xmm5
-	movdqa	%xmm1, %xmm15
-	paddd	%xmm2, %xmm14
-	movdqa	%xmm7, %xmm2
-	por	%xmm5, %xmm6
-	psrld	$16, %xmm1
-	pand	%xmm8, %xmm2
+	blendps	$0xaa, %xmm8, %xmm1
+	movaps	%xmm1, %xmm8
+	psrld	$0x10, %xmm1
+	pand	%xmm9, %xmm8
+	movaps	%xmm7, %xmm13
+	psrld	$0x10, %xmm7
+	psrlq	$0x30, %xmm13
+	pmuludq	%xmm7, %xmm6
+	pmuludq	%xmm13, %xmm4
+	psllq	$0x20, %xmm4
+	blendps	$0xaa, %xmm4, %xmm6
 	paddd	%xmm1, %xmm6
-	movdqu	160(%rsp), %xmm1
-	paddd	%xmm6, %xmm2
-	movdqu	192(%rsp), %xmm6
-	psrld	$16, %xmm7
-	pmuludq	%xmm12, %xmm1
-	pand	%xmm8, %xmm15
-	pmuludq	%xmm11, %xmm6
-	pmuludq	144(%rsp), %xmm12
-	pmuludq	176(%rsp), %xmm11
+	movaps	%xmm5, %xmm4
+	psrlq	$0x20, %xmm5
+	pmuludq	%xmm11, %xmm4
+	pmuludq	%xmm10, %xmm5
+	psllq	$0x20, %xmm5
+	blendps	$0xaa, %xmm5, %xmm4
+	pand	%xmm9, %xmm4
+	paddd	%xmm6, %xmm4
+	movaps	%xmm2, %xmm5
+	psrld	$0x10, %xmm2
+	movaps	%xmm11, %xmm6
+	pmuludq	%xmm2, %xmm11
+	pmuludq	%xmm7, %xmm2
+	movaps	%xmm5, %xmm1
+	psrlq	$0x30, %xmm5
 	pand	%xmm9, %xmm1
-	psllq	$32, %xmm6
-	por	%xmm6, %xmm1
-	psrld	$16, %xmm0
-	paddd	%xmm7, %xmm1
-	paddd	%xmm14, %xmm15
-	movdqu	128(%rsp), %xmm7
-	paddd	%xmm15, %xmm0
-	pmuludq	%xmm7, %xmm13
-	psrlq	$32, %xmm7
-	pmuludq	%xmm7, %xmm10
-	movdqa	%xmm0, %xmm14
-	pand	%xmm9, %xmm13
-	movdqu	208(%rsp), %xmm5
-	psrld	$16, %xmm14
-	paddd	%xmm2, %xmm14
-	movdqa	%xmm5, %xmm15
-	movdqa	%xmm14, %xmm3
-	pand	%xmm8, %xmm15
-	psrld	$16, %xmm3
-	paddd	%xmm1, %xmm15
-	psllq	$32, %xmm10
-	pand	%xmm9, %xmm12
-	psllq	$32, %xmm11
-	paddd	%xmm15, %xmm3
-	por	%xmm10, %xmm13
-	por	%xmm11, %xmm12
-	psrld	$16, %xmm5
-	movdqa	%xmm3, %xmm4
-	pand	%xmm8, %xmm13
+	movaps	%xmm10, %xmm12
+	pmuludq	%xmm5, %xmm10
+	psllq	$0x20, %xmm10
+	blendps	$0xaa, %xmm10, %xmm11
+	pmuludq	%xmm13, %xmm5
+	psllq	$0x20, %xmm5
+	blendps	$0xaa, %xmm5, %xmm2
+	movaps	%xmm11, %xmm5
+	pand	%xmm9, %xmm11
+	psrld	$0x10, %xmm5
+	paddd	%xmm5, %xmm2
+	paddd	%xmm2, %xmm8
+	movaps	%xmm6, %xmm5
+	pmuludq	%xmm1, %xmm6
+	movaps	%xmm1, %xmm2
+	psrlq	$0x20, %xmm1
+	pmuludq	%xmm7, %xmm2
+	movaps	%xmm12, %xmm10
+	pmuludq	%xmm1, %xmm12
+	psllq	$0x20, %xmm12
+	pmuludq	%xmm13, %xmm1
+	psllq	$0x20, %xmm1
+	blendps	$0xaa, %xmm1, %xmm2
+	blendps	$0xaa, %xmm12, %xmm6
+	movaps	%xmm6, %xmm1
+	psrld	$0x10, %xmm6
+	pand	%xmm9, %xmm1
+	paddd	%xmm6, %xmm2
+	paddd	%xmm2, %xmm11
+	movd	8(%rax, %rcx, 4), %xmm2
+	movd	8(%rax, %rdi, 4), %xmm6
+	punpckldq %xmm2, %xmm6
+	movd	8(%rax, %rdx, 4), %xmm2
+	movd	8(%rax, %rsi, 4), %xmm12
+	punpckldq %xmm2, %xmm12
+	punpcklqdq %xmm6, %xmm12
+	movaps	%xmm12, %xmm2
+	psrld	$0x10, %xmm12
+	pmuludq	%xmm12, %xmm5
+	pmuludq	%xmm7, %xmm12
+	movaps	%xmm2, %xmm6
+	psrlq	$0x30, %xmm2
+	pand	%xmm9, %xmm6
+	pmuludq	%xmm6, %xmm7
+	psrlq	$0x20, %xmm6
+	pmuludq	%xmm13, %xmm6
+	psllq	$0x20, %xmm6
+	blendps	$0xaa, %xmm6, %xmm7
+	psrld	$0x10, %xmm7
+	pmuludq	%xmm2, %xmm13
+	pmuludq	%xmm10, %xmm2
+	psllq	$0x20, %xmm2
+	psllq	$0x20, %xmm13
+	blendps	$0xaa, %xmm2, %xmm5
+	psrld	$0x10, %xmm5
+	blendps	$0xaa, %xmm13, %xmm12
 	paddd	%xmm5, %xmm12
-	psrld	$16, %xmm4
-	paddd	%xmm12, %xmm13
-	paddd	%xmm13, %xmm4
-	pand	%xmm8, %xmm3
-	pslld	$16, %xmm4
-	movd	%edx, %xmm9
-	movups	48(%rsp), %xmm15
-	paddd	%xmm3, %xmm4
-	pshufd	$0, %xmm9, %xmm7
-
-	/* Assemble reduced argument from the pieces */
-	pand	%xmm8, %xmm0
-	movd	%ecx, %xmm8
-	pand	%xmm15, %xmm7
-	pshufd	$0, %xmm8, %xmm1
-	movdqa	%xmm4, %xmm5
-	psrld	$9, %xmm5
-	pxor	%xmm7, %xmm1
-	por	%xmm1, %xmm5
-	movd	%esi, %xmm6
-	pshufd	$0, %xmm6, %xmm3
-	movdqa	%xmm5, %xmm6
-	movl	$262143, %r8d
-
-	/*
-	 * 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.
-	 */
-	movl	$679477248, %edi
-	movd	%r10d, %xmm13
-	pslld	$16, %xmm14
-	pshufd	$0, %xmm13, %xmm1
-	paddd	%xmm0, %xmm14
-	movd	%r9d, %xmm11
-	pand	%xmm4, %xmm1
-	movd	%r8d, %xmm9
-	movd	%edi, %xmm10
-	pshufd	$0, %xmm9, %xmm8
-	pslld	$14, %xmm1
-	pshufd	$0, %xmm10, %xmm0
-	pand	%xmm14, %xmm8
-	pshufd	$0, %xmm11, %xmm12
-	psrld	$18, %xmm14
-	pxor	%xmm7, %xmm0
-	pxor	%xmm12, %xmm7
-	por	%xmm14, %xmm1
-	pslld	$5, %xmm8
-	por	%xmm7, %xmm1
-
-	/*
-	 * Now multiply those numbers all by 2 pi, reasonably accurately.
-	 * The top part uses 2pi = s2pi_lead + s2pi_trail, where
-	 * s2pi_lead has 12 significant bits.
-	 */
-	movl	$1086918619, %r11d
-
-	/* Split RHi into 12-bit leading and trailing parts. */
-	movl	$-4096, %esi
-	por	%xmm0, %xmm8
-	movl	$1086918656, %edx
-	movl	$-1214941318, %ecx
-
-	/*
-	 * 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)
-	 */
-	movl	$2147483647, %edi
-	addps	%xmm3, %xmm6
-	subps	%xmm7, %xmm1
-	subps	%xmm0, %xmm8
-	movaps	%xmm6, %xmm2
-	movd	%r11d, %xmm14
-	movd	%esi, %xmm4
-	movd	%edx, %xmm7
-	movl	$897581056, %r8d
-	subps	%xmm3, %xmm2
-
-	/* Grab our final N value as an integer, appropriately masked mod 2^8 */
-	movl	$255, %r9d
-	subps	%xmm2, %xmm5
-
-	/* Now add them up into 2 reasonably aligned pieces */
-	movaps	%xmm5, %xmm3
-
-	/*
-	 * 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.
-	 */
-	movl	$127, %r10d
-	pshufd	$0, %xmm14, %xmm2
-	addps	%xmm1, %xmm3
-	pshufd	$0, %xmm4, %xmm14
-	movd	%r8d, %xmm4
-	pshufd	$0, %xmm4, %xmm9
-	subps	%xmm3, %xmm5
-	movdqa	%xmm9, %xmm11
-	addps	%xmm5, %xmm1
-	movd	%ecx, %xmm5
-	addps	%xmm1, %xmm8
-	pshufd	$0, %xmm7, %xmm1
-	movdqa	%xmm14, %xmm7
-	andps	%xmm3, %xmm7
+	paddd	%xmm12, %xmm1
+	paddd	%xmm1, %xmm7
+	movaps	%xmm7, %xmm5
+	psrld	$0x10, %xmm7
+	pand	%xmm9, %xmm5
+	paddd	%xmm11, %xmm7
+	movaps	%xmm7, %xmm2
+	psrld	$0x10, %xmm7
+	paddd	%xmm8, %xmm7
+	pslld	$0x10, %xmm2
+	paddd	%xmm5, %xmm2
+	pand	%xmm7, %xmm9
+	psrld	$0x10, %xmm7
+	paddd	%xmm4, %xmm7
+	pslld	$0x10, %xmm7
+	paddd	%xmm9, %xmm7
+	movaps	%xmm7, %xmm4
+	/* Assemble reduced argument from the pieces.  */
+	psrld	$0x9, %xmm7
+	/* Create floating-point high part, implicitly adding integer
+	   bit 1
+	   Incorporate overall sign at this stage too.  */
+	por	COMMON_DATA(_OneF)(%rip), %xmm7
+	movaps	AVX2_SHARED_DATA(_SH_FLT_1)(%rip), %xmm9
+	movaps	%xmm7, %xmm5
+	addps	%xmm9, %xmm7
+	movaps	%xmm7, %xmm6
+	subps	%xmm9, %xmm7
+	/* Grab our final N value as an integer, appropriately masked
+	   mod 2^8.  */
+	subps	%xmm7, %xmm5
+
+	movaps	%xmm2, %xmm9
+	psrld	$0x12, %xmm2
+	movaps	AVX2_SHARED_DATA(_Low9)(%rip), %xmm7
+	pand	%xmm4, %xmm7
+	pslld	$0xe, %xmm7
+	por	%xmm2, %xmm7
+	movaps	AVX2_SHARED_DATA(_SH_FLT_3)(%rip), %xmm4
+	por	%xmm4, %xmm7
+	subps	%xmm4, %xmm7
+	movaps	%xmm5, %xmm4
+	addps	%xmm7, %xmm5
 
-	/*
-	 * Do the multiplication as exact top part and "naive" low part.
-	 * This still maintains a similar level of offset and doesn't drop
-	 * the accuracy much below what we already have.
-	 */
-	movdqa	%xmm1, %xmm10
-	pshufd	$0, %xmm5, %xmm5
-	subps	%xmm7, %xmm3
-	mulps	%xmm7, %xmm10
+	/* Split RHi into 12-bit leading and trailing parts.  */
+	movaps	COMMON_DATA(_Neg4096)(%rip), %xmm0
+	subps	%xmm5, %xmm4
+	addps	%xmm4, %xmm7
+	movaps	%xmm0, %xmm4
+	andps	%xmm5, %xmm0
+	subps	%xmm0, %xmm5
+	/* Do the multiplication as exact top part and "naive" low.  */
+	movaps	LOCAL_DATA(_FLT_0)(%rip), %xmm2
+	movaps	%xmm2, %xmm8
+	mulps	%xmm5, %xmm2
+	movaps	AVX2_SHARED_DATA(_Low18)(%rip), %xmm10
+
+	mulps	%xmm0, %xmm8
+
+
+	pand	%xmm9, %xmm10
+	pslld	$0x5, %xmm10
+	movaps	AVX2_SHARED_DATA(_SH_FLT_2)(%rip), %xmm1
+
+	/* 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).  */
+
+	por	%xmm1, %xmm10
+	subps	%xmm1, %xmm10
+	addps	%xmm7, %xmm10
+
+	/* Now multiply those numbers all by 2 pi, reasonably accurately.
+	   The top part uses 2pi = s2pi_lead + s2pi_trail, where
+	   s2pi_lead has 12 significant bits.  */
+	movaps	AVX2_SHARED_DATA(_SH_FLT_4)(%rip), %xmm9
+	mulps	%xmm10, %xmm9
+	addps	%xmm2, %xmm9
+	/* Now add them up into 2 reasonably aligned pieces.  */
+	movaps	LOCAL_DATA(_FLT_1)(%rip), %xmm7
+	mulps	%xmm7, %xmm0
 	mulps	%xmm5, %xmm7
-	mulps	%xmm3, %xmm1
-	mulps	%xmm8, %xmm2
-	mulps	%xmm3, %xmm5
-	addps	%xmm7, %xmm1
-	addps	%xmm5, %xmm2
-	movd	%edi, %xmm8
-	addps	%xmm2, %xmm1
-
-	/*
-	 * Do another stage of compensated summation to get full offset
-	 * between the pieces sRedHi + sRedLo.
-	 * Depending on the later algorithm, we might avoid this stage.
-	 */
+	addps	%xmm8, %xmm7
+	addps	%xmm9, %xmm7
+	addps	%xmm7, %xmm0
+	lea	AVX2_SHARED_DATA(_Coeffs)(%rip), %rax
+
+	/* 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.  */
+	pand	AVX2_SHARED_DATA(_Low7)(%rip), %xmm6
+	movaps	%xmm6, %xmm9
+	/* Simply combine the two parts of the reduced argument
+	   since we can afford a few ulps in this case.  */
+	pslld	$0x2, %xmm6
+	paddd	%xmm9, %xmm6
+	movq	%xmm6, %rcx
+	movl	%ecx, %edx
+	shrq	$0x20, %rcx
+	pextrq	$0x1, %xmm6, %rsi
+	movl	%esi, %edi
+	shrq	$0x20, %rsi
+	movups	16(%rax, %rcx, 8), %xmm9
+	movups	16(%rax, %rdx, 8), %xmm7
+	movaps	%xmm7, %xmm5
+	punpckhdq %xmm9, %xmm7
+	punpckldq %xmm9, %xmm5
+	movups	16(%rax, %rsi, 8), %xmm9
+	movups	16(%rax, %rdi, 8), %xmm2
+	movaps	%xmm2, %xmm6
+	punpckhdq %xmm9, %xmm2
+	punpckldq %xmm9, %xmm6
+	movaps	%xmm7, %xmm9
+	punpckhqdq %xmm2, %xmm7
+	punpcklqdq %xmm2, %xmm9
+
+	/* Higher polynomial terms
+	   Stage 1 (with unlimited parallelism)
+	   P3 = C1_lo + C2 * Z.  */
+	mulps	%xmm0, %xmm7
+	addps	%xmm7, %xmm9
+	movq	32(%rax, %rsi, 8), %xmm7
+	movq	32(%rax, %rdi, 8), %xmm2
+	punpckldq %xmm7, %xmm2
+	movq	32(%rax, %rcx, 8), %xmm7
+	movq	32(%rax, %rdx, 8), %xmm8
+	punpckldq %xmm7, %xmm8
+	movaps	%xmm8, %xmm7
+	punpckhqdq %xmm2, %xmm8
+	punpcklqdq %xmm2, %xmm7
+	mulps	%xmm0, %xmm8
+	addps	%xmm8, %xmm7
+	movaps	%xmm0, %xmm2
+	mulps	%xmm0, %xmm0
+
+	mulps	%xmm0, %xmm7
+	addps	%xmm7, %xmm9
+	/* Final accumulation of low part.  */
+	mulps	%xmm2, %xmm9
+	movups	0(%rax, %rsi, 8), %xmm0
+	movups	0(%rax, %rdi, 8), %xmm7
+	movaps	%xmm7, %xmm8
+	punpckldq %xmm0, %xmm7
+	punpckhdq %xmm0, %xmm8
+	movups	0(%rax, %rcx, 8), %xmm0
+	movups	0(%rax, %rdx, 8), %xmm1
+	movaps	%xmm1, %xmm10
+	punpckldq %xmm0, %xmm1
+	punpckhdq %xmm0, %xmm10
 	movaps	%xmm1, %xmm0
-
-	/*  Load constants (not all needed at once)  */
-	lea	_sCoeffs+36+__svml_stan_data_internal(%rip), %rdi
-	pshufd	$0, %xmm8, %xmm8
-	addps	%xmm10, %xmm0
-	andps	%xmm15, %xmm8
-	subps	%xmm0, %xmm10
-	cmpltps	%xmm8, %xmm11
-	cmpleps	%xmm9, %xmm8
-	addps	%xmm10, %xmm1
-	andps	%xmm15, %xmm8
-	movd	%r9d, %xmm15
-	andps	%xmm11, %xmm0
-	andps	%xmm1, %xmm11
-	pshufd	$0, %xmm15, %xmm1
-	movd	%r10d, %xmm15
-	pshufd	$0, %xmm15, %xmm7
-	pand	%xmm1, %xmm6
-	pand	%xmm7, %xmm6
-	orps	%xmm0, %xmm8
+	punpcklqdq %xmm7, %xmm1
+	punpckhqdq %xmm7, %xmm0
+
+	/* 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.  */
+	subps	%xmm2, %xmm1
+	movaps	%xmm4, %xmm7
+	andps	%xmm1, %xmm4
+	subps	%xmm4, %xmm1
+	addps	%xmm1, %xmm0
+
+	/* 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.  */
+	rcpps	%xmm4, %xmm1
+	andps	%xmm7, %xmm1
+	mulps	%xmm1, %xmm4
+	movaps	%xmm10, %xmm7
+	punpcklqdq %xmm8, %xmm10
+	punpckhqdq %xmm8, %xmm7
+	movaps	%xmm1, %xmm8
+	/* Finally, multiplex both parts so they are only used in
+	   cotangent path.  */
+	mulps	%xmm10, %xmm1
+	movaps	%xmm5, %xmm11
+	punpckhqdq %xmm6, %xmm5
+	punpcklqdq %xmm6, %xmm11
+
+	/* Compensated sum of dominant component(s) Compute C0_hi +
+	   C1_hi * Z + Recip_hi + Recip_lo = H4 (hi) + H9 (lo) H1 =
+	   C1_hi * Z (exact since C1_hi is 1 bit).  */
+	mulps	%xmm2, %xmm5
+	movaps	%xmm7, %xmm2
+	/* H2 = high(C0_hi + C1_hi * Z).  */
+	addps	%xmm5, %xmm7
+	/* H4 = high(H2 + Recip_hi).  */
+
+	subps	%xmm7, %xmm2
+	/* H5 = low(C0_hi + C1_hi * Z).  */
+	addps	%xmm2, %xmm5
+	movaps	%xmm7, %xmm2
+	addps	%xmm1, %xmm7
+
+	/* intermediate in compensated sum.  */
+	subps	%xmm7, %xmm1
+	/* H8 = low(H2 + Recip_hi).  */
+	addps	%xmm1, %xmm2
+
+	/* Get a better approximation to  1/sR_hi (not far short of an ulp)
+	   using a third-order polynomial approximation.  */
+	movups	COMMON_DATA(_OneF)(%rip), %xmm6
+	movaps	%xmm6, %xmm1
+	subps	%xmm4, %xmm6
 	movaps	%xmm6, %xmm4
-
-	/*
-	 * Simply combine the two parts of the reduced argument
-	 * since we can afford a few ulps in this case.
-	 */
-	addps	%xmm11, %xmm8
-	pslld	$2, %xmm4
-	paddd	%xmm6, %xmm4
-	pslld	$3, %xmm4
-	pshufd	$1, %xmm4, %xmm6
-	pshufd	$2, %xmm4, %xmm5
-	pshufd	$3, %xmm4, %xmm3
-	movd	%xmm4, %r11d
-	movd	%xmm6, %edx
-	movd	%xmm5, %ecx
-	movd	%xmm3, %esi
-	movd	-32(%r11, %rdi), %xmm15
-	movd	-32(%rdx, %rdi), %xmm12
-	movd	-32(%rcx, %rdi), %xmm7
-	movd	-32(%rsi, %rdi), %xmm13
-	punpckldq %xmm12, %xmm15
-	punpckldq %xmm13, %xmm7
-	movd	-28(%rsi, %rdi), %xmm5
-	punpcklqdq %xmm7, %xmm15
-	movd	-28(%r11, %rdi), %xmm7
-	movd	-28(%rdx, %rdi), %xmm6
-	movd	-28(%rcx, %rdi), %xmm4
-	movd	-36(%rcx, %rdi), %xmm9
-	movd	-36(%r11, %rdi), %xmm1
-	movd	-36(%rdx, %rdi), %xmm2
-	movd	-24(%rdx, %rdi), %xmm3
-	movd	-36(%rsi, %rdi), %xmm10
-	punpckldq %xmm6, %xmm7
-	punpckldq %xmm5, %xmm4
-	movd	-24(%r11, %rdi), %xmm6
-	punpckldq %xmm2, %xmm1
-	punpckldq %xmm10, %xmm9
-	punpcklqdq %xmm4, %xmm7
-	movd	-16(%r11, %rdi), %xmm4
-	punpckldq %xmm3, %xmm6
-	movd	-24(%rcx, %rdi), %xmm10
-	movd	-16(%rcx, %rdi), %xmm3
-	movd	-24(%rsi, %rdi), %xmm2
-	movd	-16(%rsi, %rdi), %xmm13
-	movd	-16(%rdx, %rdi), %xmm12
-	punpcklqdq %xmm9, %xmm1
-	movd	-20(%rdx, %rdi), %xmm9
-	punpckldq %xmm2, %xmm10
-	movd	-20(%r11, %rdi), %xmm5
-	movd	-20(%rcx, %rdi), %xmm11
-	movd	-20(%rsi, %rdi), %xmm0
-	punpckldq %xmm12, %xmm4
-	punpckldq %xmm13, %xmm3
-	punpcklqdq %xmm10, %xmm6
-	movd	-12(%rsi, %rdi), %xmm10
-	punpckldq %xmm9, %xmm5
-	punpckldq %xmm0, %xmm11
-	punpcklqdq %xmm3, %xmm4
-	movd	-12(%r11, %rdi), %xmm3
-	movd	-12(%rdx, %rdi), %xmm2
-	movd	-12(%rcx, %rdi), %xmm9
-	punpcklqdq %xmm11, %xmm5
-	punpckldq %xmm2, %xmm3
-	punpckldq %xmm10, %xmm9
-	movd	-8(%rcx, %rdi), %xmm10
-	movd	-8(%r11, %rdi), %xmm2
-	movd	-8(%rdx, %rdi), %xmm0
-	movd	-8(%rsi, %rdi), %xmm11
-	punpckldq %xmm0, %xmm2
-	punpckldq %xmm11, %xmm10
-	movd	-4(%rsi, %rdi), %xmm13
-	punpcklqdq %xmm9, %xmm3
-	punpcklqdq %xmm10, %xmm2
-	movd	-4(%r11, %rdi), %xmm10
-	movd	-4(%rdx, %rdi), %xmm12
-	movd	-4(%rcx, %rdi), %xmm9
-	punpckldq %xmm12, %xmm10
-	punpckldq %xmm13, %xmm9
-	punpcklqdq %xmm9, %xmm10
-
-	/*
-	 *  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.
-	 */
-	movdqa	%xmm1, %xmm9
-	movd	(%r11, %rdi), %xmm13
-	subps	%xmm8, %xmm9
-	movd	(%rdx, %rdi), %xmm0
-	subps	%xmm9, %xmm1
-	punpckldq %xmm0, %xmm13
-	movdqa	%xmm14, %xmm0
-	andps	%xmm9, %xmm0
-	subps	%xmm8, %xmm1
-	subps	%xmm0, %xmm9
-	movd	(%rcx, %rdi), %xmm12
-	addps	%xmm9, %xmm15
-
-	/*
-	 * 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.
-	 */
-	rcpps	%xmm0, %xmm9
-	addps	%xmm15, %xmm1
-	andps	%xmm14, %xmm9
-	mulps	%xmm9, %xmm0
-
-	/*
-	 * Get a better approximation to  1/sR_hi (not far short of an ulp)
-	 * using a third-order polynomial approximation
-	 */
-	movaps	%xmm9, %xmm14
-	movd	(%rsi, %rdi), %xmm11
-
-	/*
-	 * Now compute the error sEr where sRecip_hi = (1/R_hi) * (1 - sEr)
-	 * so that we can compensate for it.
-	 */
-	movups	_sOne+__svml_stan_data_internal(%rip), %xmm15
-	punpckldq %xmm11, %xmm12
-	movaps	%xmm15, %xmm11
-	punpcklqdq %xmm12, %xmm13
-	subps	%xmm0, %xmm11
-	mulps	%xmm11, %xmm14
-	movups	%xmm11, (%rsp)
-	addps	%xmm9, %xmm14
-	mulps	%xmm11, %xmm11
-	movups	%xmm13, 32(%rsp)
-	movups	%xmm11, 16(%rsp)
-	movups	112(%rsp), %xmm0
-	movups	96(%rsp), %xmm11
-	movups	80(%rsp), %xmm12
-	movups	64(%rsp), %xmm13
-	# LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm1 xmm2 xmm3 xmm4 xmm5 xmm6 xmm7 xmm8 xmm9 xmm10 xmm11 xmm12 xmm13 xmm14 xmm15
-
-	/*
-	 *  Compensated sum of dominant component(s)
-	 * Compute C0_hi + C1_hi * Z + Recip_hi + Recip_lo = H4 (hi) + H9 (lo)
-	 * H1 = C1_hi * Z (exact since C1_hi is 1 bit)
-	 */
-	mulps	%xmm8, %xmm4
-	addps	16(%rsp), %xmm15
-
-	/* Finally, multiplex both parts so they are only used in cotangent path */
-	mulps	%xmm7, %xmm9
-
-	/*
-	 *  Higher polynomial terms
-	 * Stage 1 (with unlimited parallelism)
-	 * P3 = C1_lo + C2 * Z
-	 */
-	mulps	%xmm8, %xmm2
-	mulps	%xmm15, %xmm14
-	addps	%xmm2, %xmm3
-
-	/*
-	 * 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.
-	 */
-	mulps	%xmm14, %xmm1
-
-	/*
-	 * 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)
-	 */
-	movaps	%xmm1, %xmm15
-	mulps	%xmm1, %xmm1
-	subps	(%rsp), %xmm15
-
-	/* P4 = C3 + C4 * Z */
-	movups	32(%rsp), %xmm2
-	subps	%xmm15, %xmm1
-	mulps	%xmm8, %xmm2
-	mulps	%xmm1, %xmm14
+	mulps	%xmm6, %xmm6
+	addps	%xmm6, %xmm1
+	movaps	%xmm8, %xmm6
+	mulps	%xmm4, %xmm8
+	addps	%xmm6, %xmm8
+	mulps	%xmm1, %xmm8
+
+	/* 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.  */
+	mulps	%xmm8, %xmm0
+	movaps	%xmm0, %xmm6
+	subps	%xmm4, %xmm0
+
+	/* 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).  */
+	mulps	%xmm6, %xmm6
+	/* P4 = C3 + C4 * Z.  */
+	subps	%xmm0, %xmm6
+	mulps	%xmm6, %xmm8
+	mulps	%xmm8, %xmm10
+	/* H7 = low(C0_hi + C1_hi * Z) + Recip_lo.  */
+	addps	%xmm5, %xmm10
+	/* Z2 = Z^2.  */
+
+	/* Now H4 + H9 should be that part.  */
 	addps	%xmm2, %xmm10
-	mulps	%xmm14, %xmm7
-
-	/* H2 = high(C0_hi + C1_hi * Z) */
-	movdqa	%xmm6, %xmm14
-	addps	%xmm4, %xmm14
-
-	/* H4 = high(H2 + Recip_hi) */
-	movaps	%xmm14, %xmm1
-
-	/* intermediate in compensated sum */
-	subps	%xmm14, %xmm6
-	addps	%xmm9, %xmm1
-
-	/* H5 = low(C0_hi + C1_hi * Z) */
-	addps	%xmm6, %xmm4
-
-	/* intermediate in compensated sum */
-	subps	%xmm1, %xmm9
-
-	/* H7 = low(C0_hi + C1_hi * Z) + Recip_lo */
-	addps	%xmm4, %xmm7
-
-	/* H8 = low(H2 + Recip_hi) */
-	addps	%xmm9, %xmm14
-
-	/* Z2 = Z^2 */
-	movaps	%xmm8, %xmm4
+	/* P9 = trail(dominant part) + C0_lo.  */
+	addps	%xmm10, %xmm11
+	/* Merge results from main and large paths:.  */
+	addps	%xmm9, %xmm11
+	addps	%xmm7, %xmm11
+	/* And now the very final summation.  */
+	andps	%xmm14, %xmm11
+
+	/* The end of implementation (LA with huge args reduction)
+	   End of large arguments path (_HA_, _LA_ and _EP_).  */
+	orps	%xmm3, %xmm11
+	movups	COMMON_DATA(_AbsMask)(%rip), %xmm3
+	andnps	%xmm15, %xmm3
+
+	/* Incorperate original sign.  */
+	xorps	%xmm3, %xmm11
+	/* Return to main vector processing path.  */
+	testl	%r8d, %r8d
+	/* Go to special inputs processing branch.  */
+	jne	L(SPECIAL_VALUES_BRANCH)
+	movaps	%xmm11, %xmm0
+	ret
 
-	/* Now H4 + H9 should be that part */
-	addps	%xmm14, %xmm7
-	mulps	%xmm8, %xmm4
 
-	/* P9 = trail(dominant part) + C0_lo */
-	addps	%xmm7, %xmm5
-
-	/*
-	 * Stage 2 (with unlimited parallelism)
-	 * P6 = C1_lo + C2 * Z + C3 * Z^2 + C4 * Z^3
-	 */
-	mulps	%xmm4, %xmm10
-	addps	%xmm10, %xmm3
-
-	/* Final accumulation of low part */
-	mulps	%xmm3, %xmm8
+	/* 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):
 
-	/* Merge results from main and large paths: */
-	movaps	%xmm11, %xmm3
-	andnps	%xmm0, %xmm3
-	addps	%xmm8, %xmm5
-	movaps	%xmm3, %xmm0
+	/* Stack coming in 16-byte aligned. Set 8-byte misaligned so on
+	   call entry will be 16-byte aligned.  */
+	subq	$56, %rsp
+	cfi_def_cfa_offset (64)
+	movups	%xmm11, 24(%rsp)
+	movups	%xmm15, 40(%rsp)
+
+	/* Use rbx/rbp for callee save registers as they get short
+	   encoding for many instructions (as compared with r12/r13).  */
+	movq	%rbx, (%rsp)
+	cfi_offset (rbx, -64)
+	movq	%rbp, 8(%rsp)
+	cfi_offset (rbp, -56)
+	/* r8d has 1s where there was a special value that needs to be
+	   handled by a tanf call.  */
+	movl	%r8d, %ebx
+L(SPECIAL_VALUES_LOOP):
 
-	/* And now the very final summation */
-	addps	%xmm5, %xmm1
+	/* 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, 12] so we can restore
+	   rsp by realigning to 64. Essentially the tradeoff is 1 extra
+	   save/restore vs 2 extra instructions in the loop.  */
+	xorl	%ebp, %ebp
+	bsfl	%ebx, %ebp
 
-	/*
-	 *  The end of implementation (LA with huge args reduction)
-	 * End of large arguments path (_HA_, _LA_ and _EP_)
-	 */
+	/* Scalar math fucntion call to process special input.  */
+	movss	40(%rsp, %rbp, 4), %xmm0
+	call	tanf@PLT
 
-	pxor	%xmm12, %xmm1
-	andps	%xmm11, %xmm1
-	orps	%xmm1, %xmm0
+	/* 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.  */
+	movss	%xmm0, 24(%rsp, %rbp, 4)
+
+	leal	-1(%rbx), %eax
+	andl	%eax, %ebx
+	jnz	L(SPECIAL_VALUES_LOOP)
+
+	/* All results have been written to 24(%rsp).  */
+	movups	24(%rsp), %xmm0
+	movq	(%rsp), %rbx
+	cfi_restore (rbx)
+	movq	8(%rsp), %rbp
+	cfi_restore (rbp)
+	addq	$56, %rsp
+	cfi_def_cfa_offset (8)
+	ret
 
-	/* Return to main vector processing path */
-	jmp	L(AUX_BRANCH_RETURN)
-	# LOE rbx rbp r12 r13 r14 r15 eax xmm0 xmm13
 END(_ZGVbN4v_tanf_sse4)
 
-	.section .rodata, "a"
+	.section .rodata.sse4, "a"
 	.align	16
 
-#ifdef __svml_stan_data_internal_typedef
-typedef unsigned int VUINT32;
-typedef struct {
-	__declspec(align(16)) VUINT32 _sInvPI_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPI1_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPI2_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPI3_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPI2_ha_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPI3_ha_uisa[4][1];
-	__declspec(align(16)) VUINT32 Th_tbl_uisa[32][1];
-	__declspec(align(16)) VUINT32 Tl_tbl_uisa[32][1];
-	__declspec(align(16)) VUINT32 _sPC3_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sPC5_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sRangeReductionVal_uisa[4][1];
-	__declspec(align(16)) VUINT32 _sInvPi[4][1];
-	__declspec(align(16)) VUINT32 _sSignMask[4][1];
-	__declspec(align(16)) VUINT32 _sAbsMask[4][1];
-	__declspec(align(16)) VUINT32 _sRangeVal[4][1];
-	__declspec(align(16)) VUINT32 _sRShifter[4][1];
-	__declspec(align(16)) VUINT32 _sOne[4][1];
-	__declspec(align(16)) VUINT32 _sRangeReductionVal[4][1];
-	__declspec(align(16)) VUINT32 _sPI1[4][1];
-	__declspec(align(16)) VUINT32 _sPI2[4][1];
-	__declspec(align(16)) VUINT32 _sPI3[4][1];
-	__declspec(align(16)) VUINT32 _sPI4[4][1];
-	__declspec(align(16)) VUINT32 _sPI1_FMA[4][1];
-	__declspec(align(16)) VUINT32 _sPI2_FMA[4][1];
-	__declspec(align(16)) VUINT32 _sPI3_FMA[4][1];
-	__declspec(align(16)) VUINT32 _sP0[4][1];
-	__declspec(align(16)) VUINT32 _sP1[4][1];
-	__declspec(align(16)) VUINT32 _sQ0[4][1];
-	__declspec(align(16)) VUINT32 _sQ1[4][1];
-	__declspec(align(16)) VUINT32 _sQ2[4][1];
-	__declspec(align(16)) VUINT32 _sTwo[4][1];
-	__declspec(align(16)) VUINT32 _sCoeffs[128][10][1];
-} __svml_stan_data_internal;
-#endif
-__svml_stan_data_internal:
-	/* UISA */
-	.long	0x4122f983, 0x4122f983, 0x4122f983, 0x4122f983 /* _sInvPI_uisa */
-	.align	16
-	.long	0x3dc90fda, 0x3dc90fda, 0x3dc90fda, 0x3dc90fda /* _sPI1_uisa */
-	.align	16
-	.long	0x31a22168, 0x31a22168, 0x31a22168, 0x31a22168 /* _sPI2_uisa */
-	.align	16
-	.long	0x25c234c5, 0x25c234c5, 0x25c234c5, 0x25c234c5 /* _sPI3_uisa */
-	.align	16
-	.long	0x31a22000, 0x31a22000, 0x31a22000, 0x31a22000 /* _sPI2_ha_uisa */
-	.align	16
-	.long	0x2a34611a, 0x2a34611a, 0x2a34611a, 0x2a34611a /* _sPI3_ha_uisa */
-	/* Th_tbl_uisa for i from 0 to 31 do printsingle(tan(i*Pi/32)); */
-	.align	16
-	.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	16
-	.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	16
-	.long	0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6, 0x3eaaaaa6 /* _sPC3_uisa */
-	.align	16
-	.long	0x3e08b888, 0x3e08b888, 0x3e08b888, 0x3e08b888 /* _sPC5_uisa */
-	.align	16
-	.long	0x46010000, 0x46010000, 0x46010000, 0x46010000 /* _sRangeReductionVal_uisa */
-	.align	16
-	.long	0x3F22F983, 0x3F22F983, 0x3F22F983, 0x3F22F983 /* _sInvPi */
-	.align	16
-	.long	0x80000000, 0x80000000, 0x80000000, 0x80000000 /* _sSignMask */
-	.align	16
-	.long	0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF /* _sAbsMask */
-	.align	16
-	.long	0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 /* _sRangeVal */
-	.align	16
-	.long	0x4B400000, 0x4B400000, 0x4B400000, 0x4B400000 /* _sRShifter */
-	.align	16
-	.long	0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 /* _sOne */
-	.align	16
-	.long	0x46010000, 0x46010000, 0x46010000, 0x46010000 /* _sRangeVal */
-	.align	16
-	.long	0x3FC90000, 0x3FC90000, 0x3FC90000, 0x3FC90000 /* _sPI1 */
-	.align	16
-	.long	0x39FDA000, 0x39FDA000, 0x39FDA000, 0x39FDA000 /* _sPI2 */
-	.align	16
-	.long	0x33A22000, 0x33A22000, 0x33A22000, 0x33A22000 /* _sPI3 */
-	.align	16
-	.long	0x2C34611A, 0x2C34611A, 0x2C34611A, 0x2C34611A /* _sPI4 */
-	// PI1, PI2, and PI3 when FMA is available
-	.align	16
-	.long	0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB, 0x3FC90FDB /* _sPI1_FMA */
-	.align	16
-	.long	0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E, 0xB33BBD2E /* _sPI2_FMA */
-	.align	16
-	.long	0xA6F72CED, 0xA6F72CED, 0xA6F72CED, 0xA6F72CED /* _sPI3_FMA */
-	.align	16
-	.long	0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC /* _sP0 */
-	.align	16
-	.long	0xBDC433B4, 0xBDC433B4, 0xBDC433B4, 0xBDC433B4 /* _sP1 */
-	.align	16
-	.long	0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC, 0x3F7FFFFC /* _sQ0 */
-	.align	16
-	.long	0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB, 0xBEDBB7AB /* _sQ1 */
-	.align	16
-	.long	0x3C1F336B, 0x3C1F336B, 0x3C1F336B, 0x3C1F336B /* _sQ2 */
-	.align	16
-	.long	0x40000000, 0x40000000, 0x40000000, 0x40000000 /* _sTwo */
-	// _sCoeffs Breakpoint B = 0 * pi/128, function tan(B + x)
-	.align	16
-	.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	16
-	.type	__svml_stan_data_internal, @object
-	.size	__svml_stan_data_internal, .-__svml_stan_data_internal
-	.space	16, 0x00
-	.align	16
-
-#ifdef __svml_stan_reduction_data_internal_typedef
-typedef unsigned int VUINT32;
-typedef struct {
-	__declspec(align(16)) 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 */
-	.align	16
-	.type	__svml_stan_reduction_data_internal, @object
-	.size	__svml_stan_reduction_data_internal, .-__svml_stan_reduction_data_internal
-	.align	16
+LOCAL_DATA_NAME:
+	DATA_VEC (LOCAL_DATA_NAME, _sPI1, 0x3FC90000)
+	DATA_VEC (LOCAL_DATA_NAME, _sPI2, 0x39FDA000)
+	DATA_VEC (LOCAL_DATA_NAME, _sPI3, 0x33A22000)
+	DATA_VEC (LOCAL_DATA_NAME, _sPI4, 0x2C34611A)
+	DATA_VEC (LOCAL_DATA_NAME, _sRangeVal, 0x00800000)
+	DATA_VEC (LOCAL_DATA_NAME, _FLT_0, 0xb795777a)
+	DATA_VEC (LOCAL_DATA_NAME, _FLT_1, 0x40c91000)
 
-.FLT_16:
-	.long	0xffffffff, 0x00000000, 0xffffffff, 0x00000000
-	.type	.FLT_16, @object
-	.size	.FLT_16, 16
+	.type	LOCAL_DATA_NAME, @object
+	.size	LOCAL_DATA_NAME, .-LOCAL_DATA_NAME