AArch64: Remove SVE erf and erfc tables

By using a combination of mask-and-add instead of the shift-based
index calculation the routines can share the same table as other
variants with no performance degradation.

The tables change name because of other changes in downstream AOR.

Reviewed-by: Wilco Dijkstra  <Wilco.Dijkstra@arm.com>

16 files changed, 50 insertions, 2691 deletions

diff --git a/sysdeps/aarch64/fpu/Makefile b/sysdeps/aarch64/fpu/Makefile
index 234a6c457c..be8541f649 100644
--- a/sysdeps/aarch64/fpu/Makefile
+++ b/sysdeps/aarch64/fpu/Makefile
@@ -41,8 +41,6 @@ libmvec-support = $(addsuffix f_advsimd,$(float-advsimd-funcs)) \
                   v_log10_data \
                   erf_data \
                   erff_data \
-                  sv_erf_data \
-                  sv_erff_data \
                   v_exp_tail_data \
                   erfc_data \
                   erfcf_data \
diff --git a/sysdeps/aarch64/fpu/erf_advsimd.c b/sysdeps/aarch64/fpu/erf_advsimd.c
index c0116735e4..a48092e838 100644
--- a/sysdeps/aarch64/fpu/erf_advsimd.c
+++ b/sysdeps/aarch64/fpu/erf_advsimd.c
@@ -58,8 +58,8 @@ static inline struct entry
 lookup (uint64x2_t i)
 {
   struct entry e;
-  float64x2_t e1 = vld1q_f64 (&__erf_data.tab[vgetq_lane_u64 (i, 0)].erf),
-	      e2 = vld1q_f64 (&__erf_data.tab[vgetq_lane_u64 (i, 1)].erf);
+  float64x2_t e1 = vld1q_f64 (&__v_erf_data.tab[vgetq_lane_u64 (i, 0)].erf),
+	      e2 = vld1q_f64 (&__v_erf_data.tab[vgetq_lane_u64 (i, 1)].erf);
   e.erf = vuzp1q_f64 (e1, e2);
   e.scale = vuzp2q_f64 (e1, e2);
   return e;
diff --git a/sysdeps/aarch64/fpu/erf_data.c b/sysdeps/aarch64/fpu/erf_data.c
index 6d2dcd235c..ea01fad7ca 100644
--- a/sysdeps/aarch64/fpu/erf_data.c
+++ b/sysdeps/aarch64/fpu/erf_data.c
@@ -19,14 +19,14 @@
 
 #include "vecmath_config.h"
 
-/* Lookup table used in erf.
+/* Lookup table used in vector erf.
    For each possible rounded input r (multiples of 1/128), between
    r = 0.0 and r = 6.0 (769 values):
-   - the first entry __erff_data.tab.erf contains the values of erf(r),
-   - the second entry __erff_data.tab.scale contains the values of
+   - the first entry __v_erff_data.tab.erf contains the values of erf(r),
+   - the second entry __v_erff_data.tab.scale contains the values of
    2/sqrt(pi)*exp(-r^2). Note that indices 0 and 1 are never hit by the
    algorithm, since lookup is performed only for x >= 1/64-1/512.  */
-const struct erf_data __erf_data = {
+const struct v_erf_data __v_erf_data = {
   .tab = { { 0x0.0000000000000p+0, 0x1.20dd750429b6dp+0 },
 	   { 0x1.20dbf3deb1340p-7, 0x1.20d8f1975c85dp+0 },
 	   { 0x1.20d77083f17a0p-6, 0x1.20cb67bd452c7p+0 },
diff --git a/sysdeps/aarch64/fpu/erf_sve.c b/sysdeps/aarch64/fpu/erf_sve.c
index 7d51417406..671d55a02b 100644
--- a/sysdeps/aarch64/fpu/erf_sve.c
+++ b/sysdeps/aarch64/fpu/erf_sve.c
@@ -67,14 +67,16 @@ svfloat64_t SV_NAME_D1 (erf) (svfloat64_t x, const svbool_t pg)
   svfloat64_t a = svabs_x (pg, x);
   svfloat64_t shift = sv_f64 (dat->shift);
   svfloat64_t z = svadd_x (pg, a, shift);
-  svuint64_t i
-      = svsub_x (pg, svreinterpret_u64 (z), svreinterpret_u64 (shift));
+  svuint64_t i = svand_x (pg, svreinterpret_u64 (z), 0xfff);
+  i = svadd_x (pg, i, i);
 
   /* Lookup without shortcut for small values but with predicate to avoid
      segfault for large values and NaNs.  */
   svfloat64_t r = svsub_x (pg, z, shift);
-  svfloat64_t erfr = svld1_gather_index (a_lt_max, __sv_erf_data.erf, i);
-  svfloat64_t scale = svld1_gather_index (a_lt_max, __sv_erf_data.scale, i);
+  svfloat64_t erfr
+      = svld1_gather_index (a_lt_max, &__v_erf_data.tab[0].erf, i);
+  svfloat64_t scale
+      = svld1_gather_index (a_lt_max, &__v_erf_data.tab[0].scale, i);
 
   /* erf(x) ~ erf(r) + scale * d * poly (r, d).  */
   svfloat64_t d = svsub_x (pg, a, r);
diff --git a/sysdeps/aarch64/fpu/erfc_advsimd.c b/sysdeps/aarch64/fpu/erfc_advsimd.c
index 2f2f755c46..d05eac61a2 100644
--- a/sysdeps/aarch64/fpu/erfc_advsimd.c
+++ b/sysdeps/aarch64/fpu/erfc_advsimd.c
@@ -69,9 +69,9 @@ lookup (uint64x2_t i)
 {
   struct entry e;
   float64x2_t e1
-      = vld1q_f64 (&__erfc_data.tab[vgetq_lane_u64 (i, 0) - Off].erfc);
+      = vld1q_f64 (&__v_erfc_data.tab[vgetq_lane_u64 (i, 0) - Off].erfc);
   float64x2_t e2
-      = vld1q_f64 (&__erfc_data.tab[vgetq_lane_u64 (i, 1) - Off].erfc);
+      = vld1q_f64 (&__v_erfc_data.tab[vgetq_lane_u64 (i, 1) - Off].erfc);
   e.erfc = vuzp1q_f64 (e1, e2);
   e.scale = vuzp2q_f64 (e1, e2);
   return e;
diff --git a/sysdeps/aarch64/fpu/erfc_data.c b/sysdeps/aarch64/fpu/erfc_data.c
index 76a94e4681..8dc6a8c42c 100644
--- a/sysdeps/aarch64/fpu/erfc_data.c
+++ b/sysdeps/aarch64/fpu/erfc_data.c
@@ -19,14 +19,14 @@
 
 #include "vecmath_config.h"
 
-/* Lookup table used in erfc.
+/* Lookup table used in vector erfc.
    For each possible rounded input r (multiples of 1/128), between
    r = 0.0 and r = ~27.0 (3488 values):
-   - the first entry __erfc_data.tab.erfc contains the values of erfc(r),
-   - the second entry __erfc_data.tab.scale contains the values of
+   - the first entry __v_erfc_data.tab.erfc contains the values of erfc(r),
+   - the second entry __v_erfc_data.tab.scale contains the values of
    2/sqrt(pi)*exp(-r^2). Both values may go into subnormal range, therefore
    they are scaled by a large enough value 2^128 (fits in 8bit).  */
-const struct erfc_data __erfc_data = {
+const struct v_erfc_data __v_erfc_data = {
   .tab = { { 0x1p128, 0x1.20dd750429b6dp128 },
 	   { 0x1.fb7c9030853b3p127, 0x1.20d8f1975c85dp128 },
 	   { 0x1.f6f9447be0743p127, 0x1.20cb67bd452c7p128 },
diff --git a/sysdeps/aarch64/fpu/erfc_sve.c b/sysdeps/aarch64/fpu/erfc_sve.c
index c17d3e4484..703926ee41 100644
--- a/sysdeps/aarch64/fpu/erfc_sve.c
+++ b/sysdeps/aarch64/fpu/erfc_sve.c
@@ -104,7 +104,7 @@ svfloat64_t SV_NAME_D1 (erfc) (svfloat64_t x, const svbool_t pg)
 
   /* Lookup erfc(r) and 2/sqrt(pi)*exp(-r^2) in tables.  */
   i = svadd_x (pg, i, i);
-  const float64_t *p = &__erfc_data.tab[0].erfc - 2 * dat->off_arr;
+  const float64_t *p = &__v_erfc_data.tab[0].erfc - 2 * dat->off_arr;
   svfloat64_t erfcr = svld1_gather_index (pg, p, i);
   svfloat64_t scale = svld1_gather_index (pg, p + 1, i);
 
diff --git a/sysdeps/aarch64/fpu/erfcf_advsimd.c b/sysdeps/aarch64/fpu/erfcf_advsimd.c
index ca5bc3ab33..59b0b0d64b 100644
--- a/sysdeps/aarch64/fpu/erfcf_advsimd.c
+++ b/sysdeps/aarch64/fpu/erfcf_advsimd.c
@@ -62,13 +62,13 @@ lookup (uint32x4_t i)
 {
   struct entry e;
   float32x2_t t0
-      = vld1_f32 (&__erfcf_data.tab[vgetq_lane_u32 (i, 0) - Off].erfc);
+      = vld1_f32 (&__v_erfcf_data.tab[vgetq_lane_u32 (i, 0) - Off].erfc);
   float32x2_t t1
-      = vld1_f32 (&__erfcf_data.tab[vgetq_lane_u32 (i, 1) - Off].erfc);
+      = vld1_f32 (&__v_erfcf_data.tab[vgetq_lane_u32 (i, 1) - Off].erfc);
   float32x2_t t2
-      = vld1_f32 (&__erfcf_data.tab[vgetq_lane_u32 (i, 2) - Off].erfc);
+      = vld1_f32 (&__v_erfcf_data.tab[vgetq_lane_u32 (i, 2) - Off].erfc);
   float32x2_t t3
-      = vld1_f32 (&__erfcf_data.tab[vgetq_lane_u32 (i, 3) - Off].erfc);
+      = vld1_f32 (&__v_erfcf_data.tab[vgetq_lane_u32 (i, 3) - Off].erfc);
   float32x4_t e1 = vcombine_f32 (t0, t1);
   float32x4_t e2 = vcombine_f32 (t2, t3);
   e.erfc = vuzp1q_f32 (e1, e2);
diff --git a/sysdeps/aarch64/fpu/erfcf_data.c b/sysdeps/aarch64/fpu/erfcf_data.c
index 77fb889a78..d45087bbb9 100644
--- a/sysdeps/aarch64/fpu/erfcf_data.c
+++ b/sysdeps/aarch64/fpu/erfcf_data.c
@@ -19,14 +19,14 @@
 
 #include "vecmath_config.h"
 
-/* Lookup table used in erfcf.
+/* Lookup table used in vector erfcf.
    For each possible rounded input r (multiples of 1/64), between
    r = 0.0 and r = 10.0625 (645 values):
-   - the first entry __erfcf_data.tab.erfc contains the values of erfc(r),
-   - the second entry __erfcf_data.tab.scale contains the values of
+   - the first entry __v_erfcf_data.tab.erfc contains the values of erfc(r),
+   - the second entry __v_erfcf_data.tab.scale contains the values of
    2/sqrt(pi)*exp(-r^2). Both values may go into subnormal range, therefore
    they are scaled by a large enough value 2^47 (fits in 8 bits).  */
-const struct erfcf_data __erfcf_data = {
+const struct v_erfcf_data __v_erfcf_data = {
   .tab = { { 0x1p47, 0x1.20dd76p47 },
 	   { 0x1.f6f944p46, 0x1.20cb68p47 },
 	   { 0x1.edf3aap46, 0x1.209546p47 },
diff --git a/sysdeps/aarch64/fpu/erfcf_sve.c b/sysdeps/aarch64/fpu/erfcf_sve.c
index 48d1677eb4..ecacb933ac 100644
--- a/sysdeps/aarch64/fpu/erfcf_sve.c
+++ b/sysdeps/aarch64/fpu/erfcf_sve.c
@@ -77,7 +77,7 @@ svfloat32_t SV_NAME_F1 (erfc) (svfloat32_t x, const svbool_t pg)
 
   /* Lookup erfc(r) and 2/sqrt(pi)*exp(-r^2) in tables.  */
   i = svmul_x (pg, i, 2);
-  const float32_t *p = &__erfcf_data.tab[0].erfc - 2 * dat->off_arr;
+  const float32_t *p = &__v_erfcf_data.tab[0].erfc - 2 * dat->off_arr;
   svfloat32_t erfcr = svld1_gather_index (pg, p, i);
   svfloat32_t scale = svld1_gather_index (pg, p + 1, i);
 
diff --git a/sysdeps/aarch64/fpu/erff_advsimd.c b/sysdeps/aarch64/fpu/erff_advsimd.c
index f2fe6ff236..db39e789b6 100644
--- a/sysdeps/aarch64/fpu/erff_advsimd.c
+++ b/sysdeps/aarch64/fpu/erff_advsimd.c
@@ -47,10 +47,10 @@ static inline struct entry
 lookup (uint32x4_t i)
 {
   struct entry e;
-  float32x2_t t0 = vld1_f32 (&__erff_data.tab[vgetq_lane_u32 (i, 0)].erf);
-  float32x2_t t1 = vld1_f32 (&__erff_data.tab[vgetq_lane_u32 (i, 1)].erf);
-  float32x2_t t2 = vld1_f32 (&__erff_data.tab[vgetq_lane_u32 (i, 2)].erf);
-  float32x2_t t3 = vld1_f32 (&__erff_data.tab[vgetq_lane_u32 (i, 3)].erf);
+  float32x2_t t0 = vld1_f32 (&__v_erff_data.tab[vgetq_lane_u32 (i, 0)].erf);
+  float32x2_t t1 = vld1_f32 (&__v_erff_data.tab[vgetq_lane_u32 (i, 1)].erf);
+  float32x2_t t2 = vld1_f32 (&__v_erff_data.tab[vgetq_lane_u32 (i, 2)].erf);
+  float32x2_t t3 = vld1_f32 (&__v_erff_data.tab[vgetq_lane_u32 (i, 3)].erf);
   float32x4_t e1 = vcombine_f32 (t0, t1);
   float32x4_t e2 = vcombine_f32 (t2, t3);
   e.erf = vuzp1q_f32 (e1, e2);
diff --git a/sysdeps/aarch64/fpu/erff_data.c b/sysdeps/aarch64/fpu/erff_data.c
index 9a32940915..da38aed205 100644
--- a/sysdeps/aarch64/fpu/erff_data.c
+++ b/sysdeps/aarch64/fpu/erff_data.c
@@ -19,14 +19,14 @@
 
 #include "vecmath_config.h"
 
-/* Lookup table used in erff.
+/* Lookup table used in vector erff.
    For each possible rounded input r (multiples of 1/128), between
    r = 0.0 and r = 4.0 (513 values):
-   - the first entry __erff_data.tab.erf contains the values of erf(r),
-   - the second entry __erff_data.tab.scale contains the values of
+   - the first entry __v_erff_data.tab.erf contains the values of erf(r),
+   - the second entry __v_erff_data.tab.scale contains the values of
    2/sqrt(pi)*exp(-r^2). Note that indices 0 and 1 are never hit by the
    algorithm, since lookup is performed only for x >= 1/64-1/512.  */
-const struct erff_data __erff_data = {
+const struct v_erff_data __v_erff_data = {
   .tab = { { 0x0.000000p+0, 0x1.20dd76p+0 },
 	   { 0x1.20dbf4p-7, 0x1.20d8f2p+0 },
 	   { 0x1.20d770p-6, 0x1.20cb68p+0 },
diff --git a/sysdeps/aarch64/fpu/erff_sve.c b/sysdeps/aarch64/fpu/erff_sve.c
index 38f00db9be..0e382eb09a 100644
--- a/sysdeps/aarch64/fpu/erff_sve.c
+++ b/sysdeps/aarch64/fpu/erff_sve.c
@@ -62,18 +62,17 @@ svfloat32_t SV_NAME_F1 (erf) (svfloat32_t x, const svbool_t pg)
 
   svfloat32_t shift = sv_f32 (dat->shift);
   svfloat32_t z = svadd_x (pg, a, shift);
-  svuint32_t i
-      = svsub_x (pg, svreinterpret_u32 (z), svreinterpret_u32 (shift));
-
-  /* Saturate lookup index.  */
-  i = svsel (a_ge_max, sv_u32 (512), i);
+  svuint32_t i = svand_x (pg, svreinterpret_u32 (z), 0xfff);
+  i = svadd_x (pg, i, i);
 
   /* r and erf(r) set to 0 for |x| below min.  */
   svfloat32_t r = svsub_z (a_gt_min, z, shift);
-  svfloat32_t erfr = svld1_gather_index (a_gt_min, __sv_erff_data.erf, i);
+  svfloat32_t erfr
+      = svld1_gather_index (a_gt_min, &__v_erff_data.tab[0].erf, i);
 
   /* scale set to 2/sqrt(pi) for |x| below min.  */
-  svfloat32_t scale = svld1_gather_index (a_gt_min, __sv_erff_data.scale, i);
+  svfloat32_t scale
+      = svld1_gather_index (a_gt_min, &__v_erff_data.tab[0].scale, i);
   scale = svsel (a_gt_min, scale, sv_f32 (dat->scale));
 
   /* erf(x) ~ erf(r) + scale * d * (1 - r * d + 1/3 * d^2).  */
diff --git a/sysdeps/aarch64/fpu/sv_erf_data.c b/sysdeps/aarch64/fpu/sv_erf_data.c
deleted file mode 100644
index a53878f893..0000000000
--- a/sysdeps/aarch64/fpu/sv_erf_data.c
+++ /dev/null
@@ -1,1570 +0,0 @@
-/* Table for SVE erf approximation
-
-   Copyright (C) 2024 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <https://www.gnu.org/licenses/>.  */
-
-#include "vecmath_config.h"
-
-/* Lookup table used in vector erf.
-   For each possible rounded input r (multiples of 1/128), between
-   r = 0.0 and r = 6.0 (769 values):
-   - the first entry __erf_data.tab.erf contains the values of erf(r),
-   - the second entry __erf_data.tab.scale contains the values of
-   2/sqrt(pi)*exp(-r^2). Note that indices 0 and 1 are never hit by the
-   algorithm, since lookup is performed only for x >= 1/64-1/512.  */
-const struct sv_erf_data __sv_erf_data = {
-  .erf = { 0x0.0000000000000p+0,
-	   0x1.20dbf3deb1340p-7,
-	   0x1.20d77083f17a0p-6,
-	   0x1.b137e0cf584dcp-6,
-	   0x1.20c5645dd2538p-5,
-	   0x1.68e5d3bbc9526p-5,
-	   0x1.b0fafef135745p-5,
-	   0x1.f902a77bd3821p-5,
-	   0x1.207d480e90658p-4,
-	   0x1.44703e87e8593p-4,
-	   0x1.68591a1e83b5dp-4,
-	   0x1.8c36beb8a8d23p-4,
-	   0x1.b0081148a873ap-4,
-	   0x1.d3cbf7e70a4b3p-4,
-	   0x1.f78159ec8bb50p-4,
-	   0x1.0d939005f65e5p-3,
-	   0x1.1f5e1a35c3b89p-3,
-	   0x1.311fc15f56d14p-3,
-	   0x1.42d7fc2f64959p-3,
-	   0x1.548642321d7c6p-3,
-	   0x1.662a0bdf7a89fp-3,
-	   0x1.77c2d2a765f9ep-3,
-	   0x1.895010fdbdbfdp-3,
-	   0x1.9ad142662e14dp-3,
-	   0x1.ac45e37fe2526p-3,
-	   0x1.bdad72110a648p-3,
-	   0x1.cf076d1233237p-3,
-	   0x1.e05354b96ff36p-3,
-	   0x1.f190aa85540e2p-3,
-	   0x1.015f78a3dcf3dp-2,
-	   0x1.09eed6982b948p-2,
-	   0x1.127631eb8de32p-2,
-	   0x1.1af54e232d609p-2,
-	   0x1.236bef825d9a2p-2,
-	   0x1.2bd9db0f7827fp-2,
-	   0x1.343ed6989b7d9p-2,
-	   0x1.3c9aa8b84bedap-2,
-	   0x1.44ed18d9f6462p-2,
-	   0x1.4d35ef3e5372ep-2,
-	   0x1.5574f4ffac98ep-2,
-	   0x1.5da9f415ff23fp-2,
-	   0x1.65d4b75b00471p-2,
-	   0x1.6df50a8dff772p-2,
-	   0x1.760aba57a76bfp-2,
-	   0x1.7e15944d9d3e4p-2,
-	   0x1.861566f5fd3c0p-2,
-	   0x1.8e0a01cab516bp-2,
-	   0x1.95f3353cbb146p-2,
-	   0x1.9dd0d2b721f39p-2,
-	   0x1.a5a2aca209394p-2,
-	   0x1.ad68966569a87p-2,
-	   0x1.b522646bbda68p-2,
-	   0x1.bccfec24855b8p-2,
-	   0x1.c4710406a65fcp-2,
-	   0x1.cc058392a6d2dp-2,
-	   0x1.d38d4354c3bd0p-2,
-	   0x1.db081ce6e2a48p-2,
-	   0x1.e275eaf25e458p-2,
-	   0x1.e9d68931ae650p-2,
-	   0x1.f129d471eabb1p-2,
-	   0x1.f86faa9428f9dp-2,
-	   0x1.ffa7ea8eb5fd0p-2,
-	   0x1.03693a371519cp-1,
-	   0x1.06f794ab2cae7p-1,
-	   0x1.0a7ef5c18edd2p-1,
-	   0x1.0dff4f247f6c6p-1,
-	   0x1.1178930ada115p-1,
-	   0x1.14eab43841b55p-1,
-	   0x1.1855a5fd3dd50p-1,
-	   0x1.1bb95c3746199p-1,
-	   0x1.1f15cb50bc4dep-1,
-	   0x1.226ae840d4d70p-1,
-	   0x1.25b8a88b6dd7fp-1,
-	   0x1.28ff0240d52cdp-1,
-	   0x1.2c3debfd7d6c1p-1,
-	   0x1.2f755ce9a21f4p-1,
-	   0x1.32a54cb8db67bp-1,
-	   0x1.35cdb3a9a144dp-1,
-	   0x1.38ee8a84beb71p-1,
-	   0x1.3c07ca9cb4f9ep-1,
-	   0x1.3f196dcd0f135p-1,
-	   0x1.42236e79a5fa6p-1,
-	   0x1.4525c78dd5966p-1,
-	   0x1.4820747ba2dc2p-1,
-	   0x1.4b13713ad3513p-1,
-	   0x1.4dfeba47f63ccp-1,
-	   0x1.50e24ca35fd2cp-1,
-	   0x1.53be25d016a4fp-1,
-	   0x1.569243d2b3a9bp-1,
-	   0x1.595ea53035283p-1,
-	   0x1.5c2348ecc4dc3p-1,
-	   0x1.5ee02e8a71a53p-1,
-	   0x1.61955607dd15dp-1,
-	   0x1.6442bfdedd397p-1,
-	   0x1.66e86d0312e82p-1,
-	   0x1.69865ee075011p-1,
-	   0x1.6c1c9759d0e5fp-1,
-	   0x1.6eab18c74091bp-1,
-	   0x1.7131e5f496a5ap-1,
-	   0x1.73b1021fc0cb8p-1,
-	   0x1.762870f720c6fp-1,
-	   0x1.78983697dc96fp-1,
-	   0x1.7b00578c26037p-1,
-	   0x1.7d60d8c979f7bp-1,
-	   0x1.7fb9bfaed8078p-1,
-	   0x1.820b1202f27fbp-1,
-	   0x1.8454d5f25760dp-1,
-	   0x1.8697120d92a4ap-1,
-	   0x1.88d1cd474a2e0p-1,
-	   0x1.8b050ef253c37p-1,
-	   0x1.8d30debfc572ep-1,
-	   0x1.8f5544bd00c04p-1,
-	   0x1.91724951b8fc6p-1,
-	   0x1.9387f53df5238p-1,
-	   0x1.959651980da31p-1,
-	   0x1.979d67caa6631p-1,
-	   0x1.999d4192a5715p-1,
-	   0x1.9b95e8fd26abap-1,
-	   0x1.9d8768656cc42p-1,
-	   0x1.9f71ca72cffb6p-1,
-	   0x1.a1551a16aaeafp-1,
-	   0x1.a331628a45b92p-1,
-	   0x1.a506af4cc00f4p-1,
-	   0x1.a6d50c20fa293p-1,
-	   0x1.a89c850b7d54dp-1,
-	   0x1.aa5d265064366p-1,
-	   0x1.ac16fc7143263p-1,
-	   0x1.adca142b10f98p-1,
-	   0x1.af767a741088bp-1,
-	   0x1.b11c3c79bb424p-1,
-	   0x1.b2bb679ead19cp-1,
-	   0x1.b4540978921eep-1,
-	   0x1.b5e62fce16095p-1,
-	   0x1.b771e894d602ep-1,
-	   0x1.b8f741ef54f83p-1,
-	   0x1.ba764a2af2b78p-1,
-	   0x1.bbef0fbde6221p-1,
-	   0x1.bd61a1453ab44p-1,
-	   0x1.bece0d82d1a5cp-1,
-	   0x1.c034635b66e23p-1,
-	   0x1.c194b1d49a184p-1,
-	   0x1.c2ef0812fc1bdp-1,
-	   0x1.c443755820d64p-1,
-	   0x1.c5920900b5fd1p-1,
-	   0x1.c6dad2829ec62p-1,
-	   0x1.c81de16b14cefp-1,
-	   0x1.c95b455cce69dp-1,
-	   0x1.ca930e0e2a825p-1,
-	   0x1.cbc54b476248dp-1,
-	   0x1.ccf20ce0c0d27p-1,
-	   0x1.ce1962c0e0d8bp-1,
-	   0x1.cf3b5cdaf0c39p-1,
-	   0x1.d0580b2cfd249p-1,
-	   0x1.d16f7dbe41ca0p-1,
-	   0x1.d281c49d818d0p-1,
-	   0x1.d38eefdf64fddp-1,
-	   0x1.d4970f9ce00d9p-1,
-	   0x1.d59a33f19ed42p-1,
-	   0x1.d6986cfa798e7p-1,
-	   0x1.d791cad3eff01p-1,
-	   0x1.d8865d98abe01p-1,
-	   0x1.d97635600bb89p-1,
-	   0x1.da61623cb41e0p-1,
-	   0x1.db47f43b2980dp-1,
-	   0x1.dc29fb60715afp-1,
-	   0x1.dd0787a8bb39dp-1,
-	   0x1.dde0a90611a0dp-1,
-	   0x1.deb56f5f12d28p-1,
-	   0x1.df85ea8db188ep-1,
-	   0x1.e0522a5dfda73p-1,
-	   0x1.e11a3e8cf4eb8p-1,
-	   0x1.e1de36c75ba58p-1,
-	   0x1.e29e22a89d766p-1,
-	   0x1.e35a11b9b61cep-1,
-	   0x1.e4121370224ccp-1,
-	   0x1.e4c6372cd8927p-1,
-	   0x1.e5768c3b4a3fcp-1,
-	   0x1.e62321d06c5e0p-1,
-	   0x1.e6cc0709c8a0dp-1,
-	   0x1.e7714aec96534p-1,
-	   0x1.e812fc64db369p-1,
-	   0x1.e8b12a44944a8p-1,
-	   0x1.e94be342e6743p-1,
-	   0x1.e9e335fb56f87p-1,
-	   0x1.ea7730ed0bbb9p-1,
-	   0x1.eb07e27a133aap-1,
-	   0x1.eb9558e6b42cep-1,
-	   0x1.ec1fa258c4beap-1,
-	   0x1.eca6ccd709544p-1,
-	   0x1.ed2ae6489ac1ep-1,
-	   0x1.edabfc7453e63p-1,
-	   0x1.ee2a1d004692cp-1,
-	   0x1.eea5557137ae0p-1,
-	   0x1.ef1db32a2277cp-1,
-	   0x1.ef93436bc2daap-1,
-	   0x1.f006135426b26p-1,
-	   0x1.f0762fde45ee6p-1,
-	   0x1.f0e3a5e1a1788p-1,
-	   0x1.f14e8211e8c55p-1,
-	   0x1.f1b6d0fea5f4dp-1,
-	   0x1.f21c9f12f0677p-1,
-	   0x1.f27ff89525acfp-1,
-	   0x1.f2e0e9a6a8b09p-1,
-	   0x1.f33f7e43a706bp-1,
-	   0x1.f39bc242e43e6p-1,
-	   0x1.f3f5c1558b19ep-1,