[libclc] b52977b - [libclc] Move pow, powr & pown to the CLC library (#133294)
via cfe-commits
cfe-commits at lists.llvm.org
Fri Mar 28 01:23:28 PDT 2025
Author: Fraser Cormack
Date: 2025-03-28T08:23:24Z
New Revision: b52977b868b02625ade1f14bfbe835e299b26f0e
URL: https://github.com/llvm/llvm-project/commit/b52977b868b02625ade1f14bfbe835e299b26f0e
DIFF: https://github.com/llvm/llvm-project/commit/b52977b868b02625ade1f14bfbe835e299b26f0e.diff
LOG: [libclc] Move pow, powr & pown to the CLC library (#133294)
These functions were already nominally in the CLC library.
Similar to others, these builtins are now vectorized and are not broken
down into scalar types.
Added:
libclc/clc/include/clc/math/binary_decl_with_int_second_arg.inc
libclc/clc/include/clc/math/binary_def_with_int_second_arg.inc
libclc/clc/include/clc/math/clc_pow.h
libclc/clc/include/clc/math/clc_pown.h
libclc/clc/include/clc/math/clc_powr.h
libclc/clc/lib/generic/math/clc_pow.cl
libclc/clc/lib/generic/math/clc_pow.inc
libclc/clc/lib/generic/math/clc_pown.cl
libclc/clc/lib/generic/math/clc_pown.inc
libclc/clc/lib/generic/math/clc_powr.cl
libclc/clc/lib/generic/math/clc_powr.inc
Modified:
libclc/clc/include/clc/math/tables.h
libclc/clc/lib/generic/SOURCES
libclc/clc/lib/generic/math/clc_tables.cl
libclc/clspv/lib/SOURCES
libclc/generic/include/clc/math/pown.h
libclc/generic/lib/SOURCES
libclc/generic/lib/math/clc_exp10.cl
libclc/generic/lib/math/clc_rootn.cl
libclc/generic/lib/math/exp_helper.cl
libclc/generic/lib/math/expm1.cl
libclc/generic/lib/math/pow.cl
libclc/generic/lib/math/pown.cl
libclc/generic/lib/math/powr.cl
libclc/generic/lib/math/tables.cl
libclc/spirv/lib/SOURCES
Removed:
libclc/generic/include/clc/math/pown.inc
libclc/generic/include/math/clc_pow.h
libclc/generic/include/math/clc_pown.h
libclc/generic/include/math/clc_pown.inc
libclc/generic/include/math/clc_powr.h
libclc/generic/lib/math/clc_pow.cl
libclc/generic/lib/math/clc_pown.cl
libclc/generic/lib/math/clc_powr.cl
libclc/generic/lib/math/pown.inc
################################################################################
diff --git a/libclc/generic/include/clc/math/pown.inc b/libclc/clc/include/clc/math/binary_decl_with_int_second_arg.inc
similarity index 69%
rename from libclc/generic/include/clc/math/pown.inc
rename to libclc/clc/include/clc/math/binary_decl_with_int_second_arg.inc
index d98630f19d679..1a854abe11d7a 100644
--- a/libclc/generic/include/clc/math/pown.inc
+++ b/libclc/clc/include/clc/math/binary_decl_with_int_second_arg.inc
@@ -6,4 +6,7 @@
//
//===----------------------------------------------------------------------===//
-_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE pown(__CLC_GENTYPE a, __CLC_INTN b);
+#include <clc/utils.h>
+
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION(__CLC_GENTYPE x,
+ __CLC_INTN y);
diff --git a/libclc/generic/lib/math/pown.inc b/libclc/clc/include/clc/math/binary_def_with_int_second_arg.inc
similarity index 62%
rename from libclc/generic/lib/math/pown.inc
rename to libclc/clc/include/clc/math/binary_def_with_int_second_arg.inc
index df85226e32dcf..756bce6538ab4 100644
--- a/libclc/generic/lib/math/pown.inc
+++ b/libclc/clc/include/clc/math/binary_def_with_int_second_arg.inc
@@ -6,6 +6,12 @@
//
//===----------------------------------------------------------------------===//
-_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE pown(__CLC_GENTYPE x, __CLC_INTN y) {
- return __clc_pown(x, y);
+#include <clc/utils.h>
+
+#ifndef __CLC_FUNCTION
+#define __CLC_FUNCTION(x) __CLC_CONCAT(__clc_, x)
+#endif
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE FUNCTION(__CLC_GENTYPE x, __CLC_INTN y) {
+ return __CLC_FUNCTION(FUNCTION)(x, y);
}
diff --git a/libclc/generic/include/math/clc_pow.h b/libclc/clc/include/clc/math/clc_pow.h
similarity index 76%
rename from libclc/generic/include/math/clc_pow.h
rename to libclc/clc/include/clc/math/clc_pow.h
index ec71d98bd04e3..3a1990d5ce091 100644
--- a/libclc/generic/include/math/clc_pow.h
+++ b/libclc/clc/include/clc/math/clc_pow.h
@@ -6,8 +6,15 @@
//
//===----------------------------------------------------------------------===//
+#ifndef __CLC_MATH_CLC_POW_H__
+#define __CLC_MATH_CLC_POW_H__
+
+#define __CLC_BODY <clc/shared/binary_decl.inc>
#define __CLC_FUNCTION __clc_pow
-#define __CLC_BODY <clc/math/binary_decl_tt.inc>
+
#include <clc/math/gentype.inc>
+
#undef __CLC_BODY
#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_POW_H__
diff --git a/libclc/generic/include/math/clc_pown.h b/libclc/clc/include/clc/math/clc_pown.h
similarity index 65%
rename from libclc/generic/include/math/clc_pown.h
rename to libclc/clc/include/clc/math/clc_pown.h
index 7549de3edfbac..2b6d3220e7e58 100644
--- a/libclc/generic/include/math/clc_pown.h
+++ b/libclc/clc/include/clc/math/clc_pown.h
@@ -6,6 +6,15 @@
//
//===----------------------------------------------------------------------===//
-#define __CLC_BODY <math/clc_pown.inc>
+#ifndef __CLC_MATH_CLC_POWN_H__
+#define __CLC_MATH_CLC_POWN_H__
+
+#define __CLC_BODY <clc/math/binary_decl_with_int_second_arg.inc>
+#define __CLC_FUNCTION __clc_pown
+
#include <clc/math/gentype.inc>
+
#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_POWN_H__
diff --git a/libclc/generic/include/math/clc_powr.h b/libclc/clc/include/clc/math/clc_powr.h
similarity index 76%
rename from libclc/generic/include/math/clc_powr.h
rename to libclc/clc/include/clc/math/clc_powr.h
index 33213c5eb6c55..61ede783fe672 100644
--- a/libclc/generic/include/math/clc_powr.h
+++ b/libclc/clc/include/clc/math/clc_powr.h
@@ -6,8 +6,15 @@
//
//===----------------------------------------------------------------------===//
+#ifndef __CLC_MATH_CLC_POWR_H__
+#define __CLC_MATH_CLC_POWR_H__
+
+#define __CLC_BODY <clc/shared/binary_decl.inc>
#define __CLC_FUNCTION __clc_powr
-#define __CLC_BODY <clc/math/binary_decl_tt.inc>
+
#include <clc/math/gentype.inc>
+
#undef __CLC_BODY
#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_POWR_H__
diff --git a/libclc/clc/include/clc/math/tables.h b/libclc/clc/include/clc/math/tables.h
index 6a0cd80f9c8cb..f2118082b1bdb 100644
--- a/libclc/clc/include/clc/math/tables.h
+++ b/libclc/clc/include/clc/math/tables.h
@@ -59,18 +59,20 @@
#define USE_TABLE(NAME, IDX) TABLE_MANGLE(NAME)(IDX)
-TABLE_FUNCTION_DECL(float2, log_inv_tbl_ep);
TABLE_FUNCTION_DECL(float2, log2_tbl);
TABLE_FUNCTION_DECL(float2, log10_tbl);
TABLE_FUNCTION_DECL(uint4, pibits_tbl);
TABLE_FUNCTION_DECL(float2, sinhcosh_tbl);
TABLE_FUNCTION_DECL(float2, cbrt_tbl);
TABLE_FUNCTION_DECL(float, exp_tbl);
-TABLE_FUNCTION_DECL(float2, exp_tbl_ep);
+CLC_TABLE_FUNCTION_DECL(float, log_inv_tbl_ep_head);
+CLC_TABLE_FUNCTION_DECL(float, log_inv_tbl_ep_tail);
CLC_TABLE_FUNCTION_DECL(float, loge_tbl_lo);
CLC_TABLE_FUNCTION_DECL(float, loge_tbl_hi);
CLC_TABLE_FUNCTION_DECL(float, log_inv_tbl);
+CLC_TABLE_FUNCTION_DECL(float, exp_tbl_ep_head);
+CLC_TABLE_FUNCTION_DECL(float, exp_tbl_ep_tail);
#ifdef cl_khr_fp64
@@ -81,14 +83,17 @@ CLC_TABLE_FUNCTION_DECL(double, ln_tbl_lo);
CLC_TABLE_FUNCTION_DECL(double, ln_tbl_hi);
CLC_TABLE_FUNCTION_DECL(double, atan_jby256_tbl_head);
CLC_TABLE_FUNCTION_DECL(double, atan_jby256_tbl_tail);
-TABLE_FUNCTION_DECL(double2, two_to_jby64_ep_tbl);
+CLC_TABLE_FUNCTION_DECL(double, two_to_jby64_ep_tbl_head);
+CLC_TABLE_FUNCTION_DECL(double, two_to_jby64_ep_tbl_tail);
TABLE_FUNCTION_DECL(double2, sinh_tbl);
TABLE_FUNCTION_DECL(double2, cosh_tbl);
TABLE_FUNCTION_DECL(double, cbrt_inv_tbl);
TABLE_FUNCTION_DECL(double2, cbrt_dbl_tbl);
TABLE_FUNCTION_DECL(double2, cbrt_rem_tbl);
-TABLE_FUNCTION_DECL(double2, powlog_tbl);
-TABLE_FUNCTION_DECL(double2, log_f_inv_tbl);
+CLC_TABLE_FUNCTION_DECL(double, powlog_tbl_head);
+CLC_TABLE_FUNCTION_DECL(double, powlog_tbl_tail);
+CLC_TABLE_FUNCTION_DECL(double, log_f_inv_tbl_head);
+CLC_TABLE_FUNCTION_DECL(double, log_f_inv_tbl_tail);
#endif // cl_khr_fp64
diff --git a/libclc/clc/lib/generic/SOURCES b/libclc/clc/lib/generic/SOURCES
index 39db96d758fca..3c648edd52a7c 100644
--- a/libclc/clc/lib/generic/SOURCES
+++ b/libclc/clc/lib/generic/SOURCES
@@ -46,6 +46,9 @@ math/clc_mad.cl
math/clc_modf.cl
math/clc_nan.cl
math/clc_nextafter.cl
+math/clc_pow.cl
+math/clc_pown.cl
+math/clc_powr.cl
math/clc_remainder.cl
math/clc_remquo.cl
math/clc_rint.cl
diff --git a/libclc/clc/lib/generic/math/clc_pow.cl b/libclc/clc/lib/generic/math/clc_pow.cl
new file mode 100644
index 0000000000000..3d371938e401a
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_pow.cl
@@ -0,0 +1,22 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/clcmacro.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_ldexp.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/clc_subnormal_config.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_select.h>
+
+#define __CLC_BODY <clc_pow.inc>
+#include <clc/math/gentype.inc>
diff --git a/libclc/clc/lib/generic/math/clc_pow.inc b/libclc/clc/lib/generic/math/clc_pow.inc
new file mode 100644
index 0000000000000..98e154984aaa3
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_pow.inc
@@ -0,0 +1,433 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Computes pow using log and exp
+//
+// x^y = exp(y * log(x))
+//
+// We take care not to lose precision in the intermediate steps
+//
+// When computing log, calculate it in splits:
+//
+// r = f * (p_invead + p_inv_tail)
+// r = rh + rt
+//
+// Calculate log polynomial using r, in end addition, do:
+//
+// poly = poly + ((rh-r) + rt)
+//
+// lth = -r
+// ltt = ((xexp * log2_t) - poly) + logT
+// lt = lth + ltt
+//
+// lh = (xexp * log2_h) + logH
+// l = lh + lt
+//
+// Calculate final log answer as gh and gt:
+//
+// gh = l & higher-half bits
+// gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
+//
+// yh = y & higher-half bits
+// yt = y - yh
+//
+// Before entering computation of exp:
+//
+// vs = ((yt*gt + yt*gh) + yh*gt)
+// v = vs + yh*gh
+// vt = ((yh*gh - v) + vs)
+//
+// In calculation of exp, add vt to r that is used for poly.
+//
+// At the end of exp, do
+//
+// ((((expT * poly) + expT) + expH*poly) + expH)
+//
+//===----------------------------------------------------------------------===//
+
+#if __CLC_FPSIZE == 32
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_pow(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ __CLC_GENTYPE absx = __clc_fabs(x);
+ __CLC_INTN ix = __CLC_AS_INTN(x);
+ __CLC_INTN ax = __CLC_AS_INTN(absx);
+ __CLC_INTN xpos = ix == ax;
+
+ __CLC_INTN iy = __CLC_AS_INTN(y);
+ __CLC_INTN ay = __CLC_AS_INTN(__clc_fabs(y));
+ __CLC_INTN ypos = iy == ay;
+
+ /* Extra precise log calculation
+ * First handle case that x is close to 1
+ */
+ __CLC_GENTYPE r = 1.0f - absx;
+ __CLC_INTN near1 = __clc_fabs(r) < 0x1.0p-4f;
+ __CLC_GENTYPE r2 = r * r;
+
+ /* Coefficients are just 1/3, 1/4, 1/5 and 1/6 */
+ __CLC_GENTYPE poly = __clc_mad(
+ r,
+ __clc_mad(r,
+ __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
+ 0x1.99999ap-3f),
+ 0x1.000000p-2f),
+ 0x1.555556p-2f);
+
+ poly *= r2 * r;
+
+ __CLC_GENTYPE lth_near1 = -r2 * 0.5f;
+ __CLC_GENTYPE ltt_near1 = -poly;
+ __CLC_GENTYPE lt_near1 = lth_near1 + ltt_near1;
+ __CLC_GENTYPE lh_near1 = -r;
+ __CLC_GENTYPE l_near1 = lh_near1 + lt_near1;
+
+ /* Computations for x not near 1 */
+ __CLC_INTN m = __CLC_CONVERT_INTN(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
+ __CLC_GENTYPE mf = __CLC_CONVERT_GENTYPE(m);
+ __CLC_INTN ixs = __CLC_AS_INTN(__CLC_AS_GENTYPE(ax | 0x3f800000) - 1.0f);
+ __CLC_GENTYPE mfs = __CLC_CONVERT_GENTYPE((ixs >> EXPSHIFTBITS_SP32) - 253);
+ __CLC_INTN c = m == -127;
+ __CLC_INTN ixn = c ? ixs : ax;
+ __CLC_GENTYPE mfn = c ? mfs : mf;
+
+ __CLC_INTN indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
+
+ /* F - Y */
+ __CLC_GENTYPE f = __CLC_AS_GENTYPE(0x3f000000 | indx) -
+ __CLC_AS_GENTYPE(0x3f000000 | (ixn & MANTBITS_SP32));
+
+ indx = indx >> 16;
+ __CLC_GENTYPE rh = f * USE_TABLE(log_inv_tbl_ep_head, indx);
+ __CLC_GENTYPE rt = f * USE_TABLE(log_inv_tbl_ep_tail, indx);
+ r = rh + rt;
+
+ poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
+ (r * r);
+ poly += (rh - r) + rt;
+
+ const __CLC_GENTYPE LOG2_HEAD = 0x1.62e000p-1f; /* 0.693115234 */
+ const __CLC_GENTYPE LOG2_TAIL = 0x1.0bfbe8p-15f; /* 0.0000319461833 */
+ __CLC_GENTYPE logel = USE_TABLE(loge_tbl_lo, indx);
+ __CLC_GENTYPE logeh = USE_TABLE(loge_tbl_hi, indx);
+ __CLC_GENTYPE lth = -r;
+ __CLC_GENTYPE ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + logeh;
+ __CLC_GENTYPE lt = lth + ltt;
+ __CLC_GENTYPE lh = __clc_mad(mfn, LOG2_HEAD, logel);
+ __CLC_GENTYPE l = lh + lt;
+
+ /* Select near 1 or not */
+ lth = near1 ? lth_near1 : lth;
+ ltt = near1 ? ltt_near1 : ltt;
+ lt = near1 ? lt_near1 : lt;
+ lh = near1 ? lh_near1 : lh;
+ l = near1 ? l_near1 : l;
+
+ __CLC_GENTYPE gh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(l) & 0xfffff000);
+ __CLC_GENTYPE gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
+
+ __CLC_GENTYPE yh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(iy) & 0xfffff000);
+
+ __CLC_GENTYPE yt = y - yh;
+
+ __CLC_GENTYPE ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
+ __CLC_GENTYPE ylogx = __clc_mad(yh, gh, ylogx_s);
+ __CLC_GENTYPE ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
+
+ /* Extra precise exp of ylogx */
+ /* 64/log2 : 92.332482616893657 */
+ const __CLC_GENTYPE R_64_BY_LOG2 = 0x1.715476p+6f;
+ __CLC_INTN n = __CLC_CONVERT_INTN(ylogx * R_64_BY_LOG2);
+ __CLC_GENTYPE nf = __CLC_CONVERT_GENTYPE(n);
+
+ __CLC_INTN j = n & 0x3f;
+ m = n >> 6;
+ __CLC_INTN m2 = m << EXPSHIFTBITS_SP32;
+
+ /* log2/64 lead: 0.0108032227 */
+ const __CLC_GENTYPE R_LOG2_BY_64_LD = 0x1.620000p-7f;
+ /* log2/64 tail: 0.0000272020388 */
+ const __CLC_GENTYPE R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
+ r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
+ ylogx_t;
+
+ /* Truncated Taylor series for e^r */
+ poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
+ 0x1.000000p-1f),
+ r * r, r);
+
+ __CLC_GENTYPE exp_head = USE_TABLE(exp_tbl_ep_head, j);
+ __CLC_GENTYPE exp_tail = USE_TABLE(exp_tbl_ep_tail, j);
+
+ __CLC_GENTYPE expylogx =
+ __clc_mad(exp_head, poly, __clc_mad(exp_tail, poly, exp_tail)) + exp_head;
+ __CLC_GENTYPE sexpylogx =
+ expylogx * __CLC_AS_GENTYPE((__CLC_UINTN)0x1 << (m + 149));
+ __CLC_GENTYPE texpylogx = __CLC_AS_GENTYPE(__CLC_AS_INTN(expylogx) + m2);
+ expylogx = m < -125 ? sexpylogx : texpylogx;
+
+ /* Result is +-Inf if (ylogx + ylogx_t) > 128*log2 */
+ expylogx =
+ __clc_select(expylogx, __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32),
+ ylogx > 0x1.62e430p+6f ||
+ (ylogx == 0x1.62e430p+6f && ylogx_t > -0x1.05c610p-22f));
+
+ /* Result is 0 if ylogx < -149*log2 */
+ expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
+
+ /* Classify y:
+ * inty = 0 means not an integer.
+ * inty = 1 means odd integer.
+ * inty = 2 means even integer.
+ */
+
+ __CLC_INTN yexp =
+ __CLC_CONVERT_INTN(ay >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32 + 1;
+ __CLC_INTN mask = ((__CLC_INTN)1 << (24 - yexp)) - 1;
+ __CLC_INTN yodd = ((iy >> (24 - yexp)) & 0x1) != 0;
+ __CLC_INTN inty = yodd ? 1 : 2;
+ inty = (iy & mask) != 0 ? 0 : inty;
+ inty = yexp < 1 ? 0 : inty;
+ inty = yexp > 24 ? 2 : inty;
+
+ __CLC_GENTYPE signval =
+ __CLC_AS_GENTYPE((__CLC_AS_UINTN(expylogx) ^ SIGNBIT_SP32));
+ expylogx = ((inty == 1) && !xpos) ? signval : expylogx;
+ __CLC_INTN ret = __CLC_AS_INTN(expylogx);
+
+ /* Corner case handling */
+ ret = (!xpos && (inty == 0)) ? QNANBITPATT_SP32 : ret;
+ ret = ax < 0x3f800000 && iy == (__CLC_INTN)NINFBITPATT_SP32 ? PINFBITPATT_SP32
+ : ret;
+ ret = ax > 0x3f800000 && iy == (__CLC_INTN)NINFBITPATT_SP32 ? 0 : ret;
+ ret = ax < 0x3f800000 && iy == (__CLC_INTN)PINFBITPATT_SP32 ? 0 : ret;
+ ret = ax > 0x3f800000 && iy == (__CLC_INTN)PINFBITPATT_SP32 ? PINFBITPATT_SP32
+ : ret;
+ __CLC_BIT_INTN x_is_ninf = ix == (__CLC_INTN)NINFBITPATT_SP32;
+ __CLC_BIT_INTN x_is_pinf = ix == (__CLC_INTN)PINFBITPATT_SP32;
+ __CLC_INTN xinf =
+ xpos ? (__CLC_INTN)PINFBITPATT_SP32 : (__CLC_INTN)NINFBITPATT_SP32;
+
+ ret = ((ax == 0) && !ypos && (inty == 1)) ? xinf : ret;
+ ret = ((ax == 0) && !ypos && (inty != 1)) ? PINFBITPATT_SP32 : ret;
+ __CLC_INTN xzero = xpos ? (__CLC_INTN)0 : (__CLC_INTN)0x80000000;
+ ret = ((ax == 0) && ypos && (inty == 1)) ? xzero : ret;
+ ret = ((ax == 0) && ypos && (inty != 1)) ? 0 : ret;
+ ret = ((ax == 0) && (iy == (__CLC_INTN)NINFBITPATT_SP32)) ? PINFBITPATT_SP32
+ : ret;
+ ret = (ix == (__CLC_INTN)0xbf800000 && ay == PINFBITPATT_SP32) ? 0x3f800000
+ : ret;
+ ret = (x_is_ninf && !ypos && (inty == 1)) ? (__CLC_INTN)0x80000000 : ret;
+ ret = (x_is_ninf && !ypos && (inty != 1)) ? 0 : ret;
+ ret = (x_is_ninf && ypos && (inty == 1)) ? (__CLC_INTN)NINFBITPATT_SP32 : ret;
+ ret = (x_is_ninf && ypos && (inty != 1)) ? (__CLC_INTN)PINFBITPATT_SP32 : ret;
+ ret = (x_is_pinf && !ypos) ? 0 : ret;
+ ret = (x_is_pinf && ypos) ? PINFBITPATT_SP32 : ret;
+ ret = (ax > PINFBITPATT_SP32) ? ix : ret;
+ ret = (ay > PINFBITPATT_SP32) ? iy : ret;
+ ret = ay == 0 ? 0x3f800000 : ret;
+ ret = ix == 0x3f800000 ? 0x3f800000 : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_pow(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ const __CLC_GENTYPE real_log2_tail = 5.76999904754328540596e-08;
+ const __CLC_GENTYPE real_log2_lead = 6.93147122859954833984e-01;
+
+ __CLC_LONGN ux = __CLC_AS_LONGN(x);
+ __CLC_LONGN ax = __CLC_AS_LONGN(__clc_fabs(x));
+ __CLC_BIT_INTN xpos = ax == ux;
+
+ __CLC_LONGN uy = __CLC_AS_LONGN(y);
+ __CLC_LONGN ay = __CLC_AS_LONGN(__clc_fabs(y));
+ __CLC_BIT_INTN ypos = ay == uy;
+
+ // Extended precision log
+ __CLC_GENTYPE v, vt;
+ {
+ __CLC_INTN exp = __CLC_CONVERT_INTN(ax >> 52) - 1023;
+ __CLC_INTN mask_exp_1023 = exp == (__CLC_INTN)-1023;
+ __CLC_GENTYPE xexp = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa = ax & 0x000FFFFFFFFFFFFFL;
+
+ __CLC_LONGN temp_ux =
+ __CLC_AS_LONGN(__CLC_AS_GENTYPE(0x3ff0000000000000L | mantissa) - 1.0);
+ exp = __CLC_CONVERT_INTN((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
+ __CLC_GENTYPE xexp1 = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
+
+ xexp = __CLC_CONVERT_LONGN(mask_exp_1023) ? xexp1 : xexp;
+ mantissa = __CLC_CONVERT_LONGN(mask_exp_1023) ? mantissa1 : mantissa;
+
+ __CLC_LONGN rax = (mantissa & 0x000ff00000000000) +
+ ((mantissa & 0x0000080000000000) << 1);
+ __CLC_INTN index = __CLC_CONVERT_INTN(rax >> 44);
+
+ __CLC_GENTYPE F = __CLC_AS_GENTYPE(rax | 0x3FE0000000000000L);
+ __CLC_GENTYPE Y = __CLC_AS_GENTYPE(mantissa | 0x3FE0000000000000L);
+ __CLC_GENTYPE f = F - Y;
+ __CLC_GENTYPE log_h = USE_TABLE(log_f_inv_tbl_head, index);
+ __CLC_GENTYPE log_t = USE_TABLE(log_f_inv_tbl_tail, index);
+ __CLC_GENTYPE f_inv = (log_h + log_t) * f;
+ __CLC_GENTYPE r1 =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(f_inv) & 0xfffffffff8000000L);
+ __CLC_GENTYPE r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
+ __CLC_GENTYPE r = r1 + r2;
+
+ __CLC_GENTYPE poly = __clc_fma(
+ r,
+ __clc_fma(r,
+ __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
+ 1.0 / 4.0),
+ 1.0 / 3.0);
+ poly = poly * r * r * r;
+
+ __CLC_GENTYPE hr1r1 = 0.5 * r1 * r1;
+ __CLC_GENTYPE poly0h = r1 + hr1r1;
+ __CLC_GENTYPE poly0t = r1 - poly0h + hr1r1;
+ poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
+
+ log_h = USE_TABLE(powlog_tbl_head, index);
+ log_t = USE_TABLE(powlog_tbl_tail, index);
+
+ __CLC_GENTYPE resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
+ __CLC_GENTYPE resT = resT_t - poly0h;
+ __CLC_GENTYPE resH = __clc_fma(xexp, real_log2_lead, log_h);
+ __CLC_GENTYPE resT_h = poly0h;
+
+ __CLC_GENTYPE H = resT + resH;
+ __CLC_GENTYPE H_h =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(H) & 0xfffffffff8000000L);
+ __CLC_GENTYPE T =
+ (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
+ H = H_h;
+
+ __CLC_GENTYPE y_head =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(uy) & 0xfffffffff8000000L);
+ __CLC_GENTYPE y_tail = y - y_head;
+
+ __CLC_GENTYPE temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
+ v = __clc_fma(y_head, H, temp);
+ vt = __clc_fma(y_head, H, -v) + temp;
+ }
+
+ // Now calculate exp of (v,vt)
+
+ __CLC_GENTYPE expv;
+ {
+ const __CLC_GENTYPE max_exp_arg = 709.782712893384;
+ const __CLC_GENTYPE min_exp_arg = -745.1332191019411;
+ const __CLC_GENTYPE sixtyfour_by_lnof2 = 92.33248261689366;
+ const __CLC_GENTYPE lnof2_by_64_head = 0.010830424260348081;
+ const __CLC_GENTYPE lnof2_by_64_tail = -4.359010638708991e-10;
+
+ __CLC_GENTYPE temp = v * sixtyfour_by_lnof2;
+ __CLC_INTN n = __CLC_CONVERT_INTN(temp);
+ __CLC_GENTYPE dn = __CLC_CONVERT_GENTYPE(n);
+ __CLC_INTN j = n & 0x0000003f;
+ __CLC_INTN m = n >> 6;
+
+ __CLC_GENTYPE f1 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ __CLC_GENTYPE f2 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+ __CLC_GENTYPE f = f1 + f2;
+
+ __CLC_GENTYPE r1 = __clc_fma(dn, -lnof2_by_64_head, v);
+ __CLC_GENTYPE r2 = dn * lnof2_by_64_tail;
+ __CLC_GENTYPE r = (r1 + r2) + vt;
+
+ __CLC_GENTYPE q =
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r, 1.38889490863777199667e-03,
+ 8.33336798434219616221e-03),
+ 4.16666666662260795726e-02),
+ 1.66666666665260878863e-01),
+ 5.00000000000000008883e-01);
+ q = __clc_fma(r * r, q, r);
+
+ expv = __clc_fma(f, q, f2) + f1;
+ expv = __clc_ldexp(expv, m);
+
+ expv = v > max_exp_arg ? __CLC_AS_GENTYPE((__CLC_ULONGN)0x7FF0000000000000L)
+ : expv;
+ expv = v < min_exp_arg ? 0.0 : expv;
+ }
+
+ // See whether y is an integer.
+ // inty = 0 means not an integer.
+ // inty = 1 means odd integer.
+ // inty = 2 means even integer.
+
+ __CLC_LONGN inty;
+ {
+ __CLC_INTN yexp =
+ __CLC_CONVERT_INTN(ay >> EXPSHIFTBITS_DP64) - EXPBIAS_DP64 + 1;
+ inty = __CLC_CONVERT_LONGN(yexp < 1 ? 0 : 2);
+ inty = __CLC_CONVERT_LONGN(yexp > 53) ? 2 : inty;
+ __CLC_LONGN mask = ((__CLC_LONGN)1L << (53 - yexp)) - 1L;
+ __CLC_LONGN inty1 = (((ay & ~mask) >> (53 - yexp)) & 1L) == 1L ? 1L : 2L;
+ inty1 = (ay & mask) != 0 ? 0 : inty1;
+ inty = __CLC_CONVERT_LONGN(!(yexp < 1) && !(yexp > 53)) ? inty1 : inty;
+ }
+
+ expv *= (inty == 1) && !xpos ? -1.0 : 1.0;
+
+ __CLC_LONGN ret = __CLC_AS_LONGN(expv);
+
+ // Now all the edge cases
+ __CLC_BIT_INTN x_is_ninf = ux == (__CLC_LONGN)NINFBITPATT_DP64;
+ __CLC_BIT_INTN x_is_pinf = ux == (__CLC_LONGN)PINFBITPATT_DP64;
+ __CLC_BIT_INTN y_is_ninf = uy == (__CLC_LONGN)NINFBITPATT_DP64;
+ __CLC_BIT_INTN y_is_pinf = uy == (__CLC_LONGN)PINFBITPATT_DP64;
+ ret = !xpos && (inty == 0) ? QNANBITPATT_DP64 : ret;
+ ret = ax < 0x3ff0000000000000L && y_is_ninf ? PINFBITPATT_DP64 : ret;
+ ret = ax > 0x3ff0000000000000L && y_is_ninf ? 0L : ret;
+ ret = ax < 0x3ff0000000000000L && y_is_pinf ? 0L : ret;
+ ret = ax > 0x3ff0000000000000L && y_is_pinf ? PINFBITPATT_DP64 : ret;
+ __CLC_LONGN xinf =
+ xpos ? (__CLC_LONGN)PINFBITPATT_DP64 : (__CLC_LONGN)NINFBITPATT_DP64;
+ ret = ((ax == 0L) && !ypos && (inty == 1)) ? xinf : ret;
+ ret = ((ax == 0L) && !ypos && (inty != 1)) ? PINFBITPATT_DP64 : ret;
+ __CLC_LONGN xzero = xpos ? (__CLC_LONGN)0L : (__CLC_LONGN)0x8000000000000000L;
+ ret = ((ax == 0L) && ypos && (inty == 1)) ? xzero : ret;
+ ret = ((ax == 0L) && ypos && (inty != 1)) ? 0L : ret;
+ ret = ((ax == 0L) && y_is_ninf) ? PINFBITPATT_DP64 : ret;
+ ret = ((ux == (__CLC_LONGN)0xbff0000000000000L) && (ay == PINFBITPATT_DP64))
+ ? 0x3ff0000000000000L
+ : ret;
+ ret = (x_is_ninf && !ypos && (inty == 1)) ? (__CLC_LONGN)0x8000000000000000L
+ : ret;
+ ret = (x_is_ninf && !ypos && (inty != 1)) ? 0L : ret;
+ ret =
+ (x_is_ninf && ypos && (inty == 1)) ? (__CLC_LONGN)NINFBITPATT_DP64 : ret;
+ ret =
+ (x_is_ninf && ypos && (inty != 1)) ? (__CLC_LONGN)PINFBITPATT_DP64 : ret;
+ ret = x_is_pinf && !ypos ? 0L : ret;
+ ret = x_is_pinf && ypos ? PINFBITPATT_DP64 : ret;
+ ret = ax > PINFBITPATT_DP64 ? ux : ret;
+ ret = ay > PINFBITPATT_DP64 ? uy : ret;
+ ret = ay == 0L ? 0x3ff0000000000000L : ret;
+ ret = ux == 0x3ff0000000000000L ? 0x3ff0000000000000L : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_pow(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ return __CLC_CONVERT_GENTYPE(
+ __clc_pow(__CLC_CONVERT_FLOATN(x), __CLC_CONVERT_FLOATN(y)));
+}
+
+#endif
diff --git a/libclc/clc/lib/generic/math/clc_pown.cl b/libclc/clc/lib/generic/math/clc_pown.cl
new file mode 100644
index 0000000000000..074f212a31b72
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_pown.cl
@@ -0,0 +1,22 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/clcmacro.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_ldexp.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/clc_subnormal_config.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_select.h>
+
+#define __CLC_BODY <clc_pown.inc>
+#include <clc/math/gentype.inc>
diff --git a/libclc/clc/lib/generic/math/clc_pown.inc b/libclc/clc/lib/generic/math/clc_pown.inc
new file mode 100644
index 0000000000000..8bdc407e9ac82
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_pown.inc
@@ -0,0 +1,397 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Computes pow using log and exp
+//
+// x^y = exp(y * log(x))
+//
+// We take care not to lose precision in the intermediate steps.
+//
+// When computing log, calculate it in splits:
+//
+// r = f * (p_invead + p_inv_tail)
+// r = rh + rt
+//
+// Calculate log polynomial using r, in end addition, do:
+//
+// poly = poly + ((rh-r) + rt)
+//
+// lth = -r
+// ltt = ((xexp * log2_t) - poly) + logT
+// lt = lth + ltt
+//
+// lh = (xexp * log2_h) + logH
+// l = lh + lt
+//
+// Calculate final log answer as gh and gt:
+//
+// gh = l & higher-half bits
+// gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
+//
+// yh = y & higher-half bits
+// yt = y - yh
+//
+// Before entering computation of exp:
+//
+// vs = ((yt*gt + yt*gh) + yh*gt)
+// v = vs + yh*gh
+// vt = ((yh*gh - v) + vs)
+//
+// In calculation of exp, add vt to r that is used for poly.
+//
+// At the end of exp, do:
+//
+// ((((expT * poly) + expT) + expH*poly) + expH)
+//
+//===----------------------------------------------------------------------===//
+
+#if __CLC_FPSIZE == 32
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_pown(__CLC_GENTYPE x,
+ __CLC_INTN ny) {
+ __CLC_GENTYPE y = __CLC_CONVERT_GENTYPE(ny);
+
+ __CLC_GENTYPE absx = __clc_fabs(x);
+ __CLC_INTN ix = __CLC_AS_INTN(x);
+ __CLC_INTN ax = __CLC_AS_INTN(absx);
+ __CLC_INTN xpos = ix == ax;
+
+ __CLC_INTN iy = __CLC_AS_INTN(y);
+ __CLC_INTN ay = __CLC_AS_INTN(__clc_fabs(y));
+ __CLC_INTN ypos = iy == ay;
+
+ // Extra precise log calculation
+ // First handle case that x is close to 1
+ __CLC_GENTYPE r = 1.0f - absx;
+ __CLC_INTN near1 = __clc_fabs(r) < 0x1.0p-4f;
+ __CLC_GENTYPE r2 = r * r;
+
+ // Coefficients are just 1/3, 1/4, 1/5 and 1/6
+ __CLC_GENTYPE poly = __clc_mad(
+ r,
+ __clc_mad(r,
+ __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
+ 0x1.99999ap-3f),
+ 0x1.000000p-2f),
+ 0x1.555556p-2f);
+
+ poly *= r2 * r;
+
+ __CLC_GENTYPE lth_near1 = -r2 * 0.5f;
+ __CLC_GENTYPE ltt_near1 = -poly;
+ __CLC_GENTYPE lt_near1 = lth_near1 + ltt_near1;
+ __CLC_GENTYPE lh_near1 = -r;
+ __CLC_GENTYPE l_near1 = lh_near1 + lt_near1;
+
+ // Computations for x not near 1
+ __CLC_INTN m = __CLC_CONVERT_INTN(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
+ __CLC_GENTYPE mf = __CLC_CONVERT_GENTYPE(m);
+ __CLC_INTN ixs = __CLC_AS_INTN(__CLC_AS_GENTYPE(ax | 0x3f800000) - 1.0f);
+ __CLC_GENTYPE mfs = __CLC_CONVERT_GENTYPE((ixs >> EXPSHIFTBITS_SP32) - 253);
+ __CLC_INTN c = m == -127;
+ __CLC_INTN ixn = c ? ixs : ax;
+ __CLC_GENTYPE mfn = c ? mfs : mf;
+
+ __CLC_INTN indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
+
+ // F - Y
+ __CLC_GENTYPE f = __CLC_AS_GENTYPE(0x3f000000 | indx) -
+ __CLC_AS_GENTYPE(0x3f000000 | (ixn & MANTBITS_SP32));
+
+ indx = indx >> 16;
+ __CLC_GENTYPE rh = f * USE_TABLE(log_inv_tbl_ep_head, indx);
+ __CLC_GENTYPE rt = f * USE_TABLE(log_inv_tbl_ep_tail, indx);
+ r = rh + rt;
+
+ poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
+ (r * r);
+ poly += (rh - r) + rt;
+
+ const __CLC_GENTYPE LOG2_HEAD = 0x1.62e000p-1f; // 0.693115234
+ const __CLC_GENTYPE LOG2_TAIL = 0x1.0bfbe8p-15f; // 0.0000319461833
+ __CLC_GENTYPE logel = USE_TABLE(loge_tbl_lo, indx);
+ __CLC_GENTYPE logeh = USE_TABLE(loge_tbl_hi, indx);
+ __CLC_GENTYPE lth = -r;
+ __CLC_GENTYPE ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + logeh;
+ __CLC_GENTYPE lt = lth + ltt;
+ __CLC_GENTYPE lh = __clc_mad(mfn, LOG2_HEAD, logel);
+ __CLC_GENTYPE l = lh + lt;
+
+ // Select near 1 or not
+ lth = near1 ? lth_near1 : lth;
+ ltt = near1 ? ltt_near1 : ltt;
+ lt = near1 ? lt_near1 : lt;
+ lh = near1 ? lh_near1 : lh;
+ l = near1 ? l_near1 : l;
+
+ __CLC_GENTYPE gh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(l) & 0xfffff000);
+ __CLC_GENTYPE gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
+
+ __CLC_GENTYPE yh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(iy) & 0xfffff000);
+
+ __CLC_GENTYPE yt = __CLC_CONVERT_GENTYPE(ny - __CLC_CONVERT_INTN(yh));
+
+ __CLC_GENTYPE ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
+ __CLC_GENTYPE ylogx = __clc_mad(yh, gh, ylogx_s);
+ __CLC_GENTYPE ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
+
+ // Extra precise exp of ylogx
+ // 64/log2 : 92.332482616893657
+ const __CLC_GENTYPE R_64_BY_LOG2 = 0x1.715476p+6f;
+ __CLC_INTN n = __CLC_CONVERT_INTN(ylogx * R_64_BY_LOG2);
+ __CLC_GENTYPE nf = __CLC_CONVERT_GENTYPE(n);
+
+ __CLC_INTN j = n & 0x3f;
+ m = n >> 6;
+ __CLC_INTN m2 = m << EXPSHIFTBITS_SP32;
+
+ // log2/64 lead: 0.0108032227
+ const __CLC_GENTYPE R_LOG2_BY_64_LD = 0x1.620000p-7f;
+ // log2/64 tail: 0.0000272020388
+ const __CLC_GENTYPE R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
+ r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
+ ylogx_t;
+
+ // Truncated Taylor series for e^r
+ poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
+ 0x1.000000p-1f),
+ r * r, r);
+
+ __CLC_GENTYPE exp_head = USE_TABLE(exp_tbl_ep_head, j);
+ __CLC_GENTYPE exp_tail = USE_TABLE(exp_tbl_ep_tail, j);
+
+ __CLC_GENTYPE expylogx =
+ __clc_mad(exp_head, poly, __clc_mad(exp_tail, poly, exp_tail)) + exp_head;
+ __CLC_GENTYPE sexpylogx =
+ expylogx * __CLC_AS_GENTYPE((__CLC_INTN)0x1 << (m + 149));
+ __CLC_GENTYPE texpylogx = __CLC_AS_GENTYPE(__CLC_AS_INTN(expylogx) + m2);
+ expylogx = m < -125 ? sexpylogx : texpylogx;
+
+ // Result is +-Inf if (ylogx + ylogx_t) > 128*log2
+ expylogx =
+ __clc_select(expylogx, __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32),
+ ylogx > 0x1.62e430p+6f ||
+ (ylogx == 0x1.62e430p+6f && ylogx_t > -0x1.05c610p-22f));
+
+ // Result is 0 if ylogx < -149*log2
+ expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
+
+ // Classify y:
+ // inty = 0 means not an integer.
+ // inty = 1 means odd integer.
+ // inty = 2 means even integer.
+
+ __CLC_INTN inty = 2 - (ny & 1);
+
+ __CLC_GENTYPE signval =
+ __CLC_AS_GENTYPE((__CLC_AS_UINTN(expylogx) ^ SIGNBIT_SP32));
+ expylogx = ((inty == 1) && !xpos) ? signval : expylogx;
+ __CLC_INTN ret = __CLC_AS_INTN(expylogx);
+
+ // Corner case handling
+ __CLC_BIT_INTN x_is_ninf = ix == (__CLC_INTN)NINFBITPATT_SP32;
+
+ __CLC_INTN xinf =
+ xpos ? (__CLC_INTN)PINFBITPATT_SP32 : (__CLC_INTN)NINFBITPATT_SP32;
+ ret = ((ax == 0) && !ypos && (inty == 1)) ? xinf : ret;
+ ret = ((ax == 0) && !ypos && (inty == 2)) ? PINFBITPATT_SP32 : ret;
+ ret = ((ax == 0) && ypos && (inty == 2)) ? 0 : ret;
+ __CLC_INTN xzero = !xpos ? (__CLC_INTN)0x80000000 : (__CLC_INTN)0;
+ ret = ((ax == 0) && ypos && (inty == 1)) ? xzero : ret;
+ ret = (x_is_ninf && !ypos && (inty == 1)) ? (__CLC_INTN)0x80000000 : ret;
+ ret = (x_is_ninf && !ypos && (inty != 1)) ? 0 : ret;
+ ret = (x_is_ninf && ypos && (inty == 1)) ? (__CLC_INTN)NINFBITPATT_SP32 : ret;
+ ret = (x_is_ninf && ypos && (inty != 1)) ? (__CLC_INTN)PINFBITPATT_SP32 : ret;
+ ret = ((ix == PINFBITPATT_SP32) && !ypos) ? 0 : ret;
+ ret = ((ix == PINFBITPATT_SP32) && ypos) ? (__CLC_INTN)PINFBITPATT_SP32 : ret;
+ ret = ax > PINFBITPATT_SP32 ? ix : ret;
+ ret = ny == 0 ? 0x3f800000 : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_pown(__CLC_GENTYPE x,
+ __CLC_INTN ny) {
+ const __CLC_GENTYPE real_log2_tail = 5.76999904754328540596e-08;
+ const __CLC_GENTYPE real_log2_lead = 6.93147122859954833984e-01;
+
+ __CLC_GENTYPE y = __CLC_CONVERT_GENTYPE(ny);
+
+ __CLC_LONGN ux = __CLC_AS_LONGN(x);
+ __CLC_LONGN ax = __CLC_AS_LONGN(__clc_fabs(x));
+ __CLC_BIT_INTN xpos = ax == ux;
+
+ __CLC_LONGN uy = __CLC_AS_LONGN(y);
+ __CLC_LONGN ay = __CLC_AS_LONGN(__clc_fabs(y));
+ __CLC_BIT_INTN ypos = ay == uy;
+
+ // Extended precision log
+ __CLC_GENTYPE v, vt;
+ {
+ __CLC_INTN exp = __CLC_CONVERT_INTN(ax >> 52) - 1023;
+ __CLC_INTN mask_exp_1023 = exp == -1023;
+ __CLC_GENTYPE xexp = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa = ax & 0x000FFFFFFFFFFFFFL;
+
+ __CLC_LONGN temp_ux =
+ __CLC_AS_LONGN(__CLC_AS_GENTYPE(0x3ff0000000000000L | mantissa) - 1.0);
+ exp = __CLC_CONVERT_INTN((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
+ __CLC_GENTYPE xexp1 = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
+
+ xexp = __CLC_CONVERT_LONGN(mask_exp_1023) ? xexp1 : xexp;
+ mantissa = __CLC_CONVERT_LONGN(mask_exp_1023) ? mantissa1 : mantissa;
+
+ __CLC_LONGN rax = (mantissa & 0x000ff00000000000) +
+ ((mantissa & 0x0000080000000000) << 1);
+ __CLC_INTN index = __CLC_CONVERT_INTN(rax >> 44);
+
+ __CLC_GENTYPE F = __CLC_AS_GENTYPE(rax | 0x3FE0000000000000L);
+ __CLC_GENTYPE Y = __CLC_AS_GENTYPE(mantissa | 0x3FE0000000000000L);
+ __CLC_GENTYPE f = F - Y;
+ __CLC_GENTYPE log_h = USE_TABLE(log_f_inv_tbl_head, index);
+ __CLC_GENTYPE log_t = USE_TABLE(log_f_inv_tbl_tail, index);
+ __CLC_GENTYPE f_inv = (log_h + log_t) * f;
+ __CLC_GENTYPE r1 =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(f_inv) & 0xfffffffff8000000L);
+ __CLC_GENTYPE r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
+ __CLC_GENTYPE r = r1 + r2;
+
+ __CLC_GENTYPE poly = __clc_fma(
+ r,
+ __clc_fma(r,
+ __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
+ 1.0 / 4.0),
+ 1.0 / 3.0);
+ poly = poly * r * r * r;
+
+ __CLC_GENTYPE hr1r1 = 0.5 * r1 * r1;
+ __CLC_GENTYPE poly0h = r1 + hr1r1;
+ __CLC_GENTYPE poly0t = r1 - poly0h + hr1r1;
+ poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
+
+ log_h = USE_TABLE(powlog_tbl_head, index);
+ log_t = USE_TABLE(powlog_tbl_tail, index);
+
+ __CLC_GENTYPE resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
+ __CLC_GENTYPE resT = resT_t - poly0h;
+ __CLC_GENTYPE resH = __clc_fma(xexp, real_log2_lead, log_h);
+ __CLC_GENTYPE resT_h = poly0h;
+
+ __CLC_GENTYPE H = resT + resH;
+ __CLC_GENTYPE H_h =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(H) & 0xfffffffff8000000L);
+ __CLC_GENTYPE T =
+ (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
+ H = H_h;
+
+ __CLC_GENTYPE y_head =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(uy) & 0xfffffffff8000000L);
+ __CLC_GENTYPE y_tail = y - y_head;
+
+ __CLC_BIT_INTN mask_2_24 = ay > 0x4170000000000000; // 2^24
+ __CLC_INTN nyh = __CLC_CONVERT_INTN(y_head);
+ __CLC_INTN nyt = ny - nyh;
+ __CLC_GENTYPE y_tail1 = __CLC_CONVERT_GENTYPE(nyt);
+ y_tail = mask_2_24 ? y_tail1 : y_tail;
+
+ __CLC_GENTYPE temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
+ v = __clc_fma(y_head, H, temp);
+ vt = __clc_fma(y_head, H, -v) + temp;
+ }
+
+ // Now calculate exp of (v,vt)
+
+ __CLC_GENTYPE expv;
+ {
+ const __CLC_GENTYPE max_exp_arg = 709.782712893384;
+ const __CLC_GENTYPE min_exp_arg = -745.1332191019411;
+ const __CLC_GENTYPE sixtyfour_by_lnof2 = 92.33248261689366;
+ const __CLC_GENTYPE lnof2_by_64_head = 0.010830424260348081;
+ const __CLC_GENTYPE lnof2_by_64_tail = -4.359010638708991e-10;
+
+ __CLC_GENTYPE temp = v * sixtyfour_by_lnof2;
+ __CLC_INTN n = __CLC_CONVERT_INTN(temp);
+ __CLC_GENTYPE dn = __CLC_CONVERT_GENTYPE(n);
+ __CLC_INTN j = n & 0x0000003f;
+ __CLC_INTN m = n >> 6;
+
+ __CLC_GENTYPE f1 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ __CLC_GENTYPE f2 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+ __CLC_GENTYPE f = f1 + f2;
+
+ __CLC_GENTYPE r1 = __clc_fma(dn, -lnof2_by_64_head, v);
+ __CLC_GENTYPE r2 = dn * lnof2_by_64_tail;
+ __CLC_GENTYPE r = (r1 + r2) + vt;
+
+ __CLC_GENTYPE q =
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r, 1.38889490863777199667e-03,
+ 8.33336798434219616221e-03),
+ 4.16666666662260795726e-02),
+ 1.66666666665260878863e-01),
+ 5.00000000000000008883e-01);
+ q = __clc_fma(r * r, q, r);
+
+ expv = __clc_fma(f, q, f2) + f1;
+ expv = __clc_ldexp(expv, m);
+
+ expv = v > max_exp_arg ? __CLC_AS_GENTYPE((__CLC_ULONGN)0x7FF0000000000000L)
+ : expv;
+ expv = v < min_exp_arg ? 0.0 : expv;
+ }
+
+ // See whether y is an integer.
+ // inty = 0 means not an integer.
+ // inty = 1 means odd integer.
+ // inty = 2 means even integer.
+
+ __CLC_LONGN inty = __CLC_CONVERT_LONGN(2 - (ny & 1));
+
+ expv *= ((inty == 1) && !xpos) ? -1.0 : 1.0;
+
+ __CLC_LONGN ret = __CLC_AS_LONGN(expv);
+
+ // Now all the edge cases
+ __CLC_BIT_INTN x_is_ninf = ux == (__CLC_LONGN)NINFBITPATT_DP64;
+ __CLC_BIT_INTN x_is_pinf = ux == (__CLC_LONGN)PINFBITPATT_DP64;
+ __CLC_LONGN xinf =
+ xpos ? (__CLC_LONGN)PINFBITPATT_DP64 : (__CLC_LONGN)NINFBITPATT_DP64;
+
+ ret = ((ax == 0L) && !ypos && (inty == 1)) ? xinf : ret;
+ ret = ((ax == 0L) && !ypos && (inty == 2)) ? (__CLC_LONGN)PINFBITPATT_DP64
+ : ret;
+ ret = ((ax == 0L) && ypos && (inty == 2)) ? 0L : ret;
+ __CLC_LONGN xzero = !xpos ? (__CLC_LONGN)0x8000000000000000L : 0L;
+ ret = ((ax == 0L) && ypos && (inty == 1)) ? xzero : ret;
+ ret = (x_is_ninf && !ypos && (inty == 1)) ? (__CLC_LONGN)0x8000000000000000L
+ : ret;
+ ret = (x_is_ninf && !ypos && (inty != 1)) ? 0L : ret;
+ ret =
+ (x_is_ninf && ypos && (inty == 1)) ? (__CLC_LONGN)NINFBITPATT_DP64 : ret;
+ ret =
+ (x_is_ninf && ypos && (inty != 1)) ? (__CLC_LONGN)PINFBITPATT_DP64 : ret;
+ ret = (x_is_pinf && !ypos) ? 0L : ret;
+ ret = (x_is_pinf && ypos) ? (__CLC_LONGN)PINFBITPATT_DP64 : ret;
+ ret = ax > (__CLC_LONGN)PINFBITPATT_DP64 ? ux : ret;
+ ret = __CLC_CONVERT_LONGN(ny == 0) ? (__CLC_LONGN)0x3ff0000000000000L : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_pown(__CLC_GENTYPE x, __CLC_INTN y) {
+ return __CLC_CONVERT_GENTYPE(__clc_pown(__CLC_CONVERT_FLOATN(x), y));
+}
+
+#endif
diff --git a/libclc/clc/lib/generic/math/clc_powr.cl b/libclc/clc/lib/generic/math/clc_powr.cl
new file mode 100644
index 0000000000000..c596a552f07fe
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_powr.cl
@@ -0,0 +1,22 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/clcmacro.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_ldexp.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/clc_subnormal_config.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_select.h>
+
+#define __CLC_BODY <clc_powr.inc>
+#include <clc/math/gentype.inc>
diff --git a/libclc/clc/lib/generic/math/clc_powr.inc b/libclc/clc/lib/generic/math/clc_powr.inc
new file mode 100644
index 0000000000000..fbdf3d85de2b7
--- /dev/null
+++ b/libclc/clc/lib/generic/math/clc_powr.inc
@@ -0,0 +1,409 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Computes pow using log and exp
+//
+// x^y = exp(y * log(x))
+//
+// We take care not to lose precision in the intermediate steps
+//
+// When computing log, calculate it in splits:
+//
+// r = f * (p_invead + p_inv_tail)
+// r = rh + rt
+//
+// Calculate log polynomial using r, in end addition, do:
+//
+// poly = poly + ((rh-r) + rt)
+//
+// lth = -r
+// ltt = ((xexp * log2_t) - poly) + logT
+// lt = lth + ltt
+//
+// lh = (xexp * log2_h) + logH
+// l = lh + lt
+//
+// Calculate final log answer as gh and gt:
+//
+// gh = l & higher-half bits
+// gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
+//
+// yh = y & higher-half bits
+// yt = y - yh
+//
+// Before entering computation of exp:
+//
+// vs = ((yt*gt + yt*gh) + yh*gt)
+// v = vs + yh*gh
+// vt = ((yh*gh - v) + vs)
+//
+// In calculation of exp, add vt to r that is used for poly.
+//
+// At the end of exp, do
+//
+// ((((expT * poly) + expT) + expH*poly) + expH)
+//
+//===----------------------------------------------------------------------===//
+
+#if __CLC_FPSIZE == 32
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_powr(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ __CLC_GENTYPE absx = __clc_fabs(x);
+ __CLC_INTN ix = __CLC_AS_INTN(x);
+ __CLC_INTN ax = __CLC_AS_INTN(absx);
+ __CLC_INTN xpos = ix == ax;
+
+ __CLC_INTN iy = __CLC_AS_INTN(y);
+ __CLC_INTN ay = __CLC_AS_INTN(__clc_fabs(y));
+ __CLC_INTN ypos = iy == ay;
+
+ // Extra precise log calculation
+ // First handle case that x is close to 1
+ __CLC_GENTYPE r = 1.0f - absx;
+ __CLC_INTN near1 = __clc_fabs(r) < 0x1.0p-4f;
+ __CLC_GENTYPE r2 = r * r;
+
+ // Coefficients are just 1/3, 1/4, 1/5 and 1/6
+ __CLC_GENTYPE poly = __clc_mad(
+ r,
+ __clc_mad(r,
+ __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
+ 0x1.99999ap-3f),
+ 0x1.000000p-2f),
+ 0x1.555556p-2f);
+
+ poly *= r2 * r;
+
+ __CLC_GENTYPE lth_near1 = -r2 * 0.5f;
+ __CLC_GENTYPE ltt_near1 = -poly;
+ __CLC_GENTYPE lt_near1 = lth_near1 + ltt_near1;
+ __CLC_GENTYPE lh_near1 = -r;
+ __CLC_GENTYPE l_near1 = lh_near1 + lt_near1;
+
+ // Computations for x not near 1
+ __CLC_INTN m = __CLC_CONVERT_INTN(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
+ __CLC_GENTYPE mf = __CLC_CONVERT_GENTYPE(m);
+ __CLC_INTN ixs = __CLC_AS_INTN(__CLC_AS_GENTYPE(ax | 0x3f800000) - 1.0f);
+ __CLC_GENTYPE mfs = __CLC_CONVERT_GENTYPE((ixs >> EXPSHIFTBITS_SP32) - 253);
+ __CLC_INTN c = m == -127;
+ __CLC_INTN ixn = c ? ixs : ax;
+ __CLC_GENTYPE mfn = c ? mfs : mf;
+
+ __CLC_INTN indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
+
+ // F - Y
+ __CLC_GENTYPE f = __CLC_AS_GENTYPE(0x3f000000 | indx) -
+ __CLC_AS_GENTYPE(0x3f000000 | (ixn & MANTBITS_SP32));
+
+ indx = indx >> 16;
+ __CLC_GENTYPE rh = f * USE_TABLE(log_inv_tbl_ep_head, indx);
+ __CLC_GENTYPE rt = f * USE_TABLE(log_inv_tbl_ep_tail, indx);
+ r = rh + rt;
+
+ poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
+ (r * r);
+ poly += (rh - r) + rt;
+
+ const __CLC_GENTYPE LOG2_HEAD = 0x1.62e000p-1f; // 0.693115234
+ const __CLC_GENTYPE LOG2_TAIL = 0x1.0bfbe8p-15f; // 0.0000319461833
+ __CLC_GENTYPE logel = USE_TABLE(loge_tbl_lo, indx);
+ __CLC_GENTYPE logeh = USE_TABLE(loge_tbl_hi, indx);
+ __CLC_GENTYPE lth = -r;
+ __CLC_GENTYPE ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + logeh;
+ __CLC_GENTYPE lt = lth + ltt;
+ __CLC_GENTYPE lh = __clc_mad(mfn, LOG2_HEAD, logel);
+ __CLC_GENTYPE l = lh + lt;
+
+ // Select near 1 or not
+ lth = near1 ? lth_near1 : lth;
+ ltt = near1 ? ltt_near1 : ltt;
+ lt = near1 ? lt_near1 : lt;
+ lh = near1 ? lh_near1 : lh;
+ l = near1 ? l_near1 : l;
+
+ __CLC_GENTYPE gh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(l) & 0xfffff000);
+ __CLC_GENTYPE gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
+
+ __CLC_GENTYPE yh = __CLC_AS_GENTYPE(__CLC_AS_UINTN(iy) & 0xfffff000);
+
+ __CLC_GENTYPE yt = y - yh;
+
+ __CLC_GENTYPE ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
+ __CLC_GENTYPE ylogx = __clc_mad(yh, gh, ylogx_s);
+ __CLC_GENTYPE ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
+
+ // Extra precise exp of ylogx
+ // 64/log2 : 92.332482616893657
+ const __CLC_GENTYPE R_64_BY_LOG2 = 0x1.715476p+6f;
+ __CLC_INTN n = __CLC_CONVERT_INTN(ylogx * R_64_BY_LOG2);
+ __CLC_GENTYPE nf = __CLC_CONVERT_GENTYPE(n);
+
+ __CLC_INTN j = n & 0x3f;
+ m = n >> 6;
+ __CLC_INTN m2 = m << EXPSHIFTBITS_SP32;
+ // log2/64 lead: 0.0108032227
+ const __CLC_GENTYPE R_LOG2_BY_64_LD = 0x1.620000p-7f;
+ // log2/64 tail: 0.0000272020388
+ const __CLC_GENTYPE R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
+ r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
+ ylogx_t;
+
+ // Truncated Taylor series for e^r
+ poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
+ 0x1.000000p-1f),
+ r * r, r);
+
+ __CLC_GENTYPE exp_head = USE_TABLE(exp_tbl_ep_head, j);
+ __CLC_GENTYPE exp_tail = USE_TABLE(exp_tbl_ep_tail, j);
+
+ __CLC_GENTYPE expylogx =
+ __clc_mad(exp_head, poly, __clc_mad(exp_tail, poly, exp_tail)) + exp_head;
+ __CLC_GENTYPE sexpylogx =
+ expylogx * __CLC_AS_GENTYPE((__CLC_INTN)0x1 << (m + 149));
+ __CLC_GENTYPE texpylogx = __CLC_AS_GENTYPE(__CLC_AS_INTN(expylogx) + m2);
+ expylogx = m < -125 ? sexpylogx : texpylogx;
+
+ // Result is +-Inf if (ylogx + ylogx_t) > 128*log2
+ expylogx =
+ __clc_select(expylogx, __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32),
+ (ylogx > 0x1.62e430p+6f) ||
+ (ylogx == 0x1.62e430p+6f && ylogx_t > -0x1.05c610p-22f));
+
+ // Result is 0 if ylogx < -149*log2
+ expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
+
+ // Classify y:
+ // inty = 0 means not an integer.
+ // inty = 1 means odd integer.
+ // inty = 2 means even integer.
+
+ __CLC_INTN yexp = (__CLC_INTN)(ay >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32 + 1;
+ __CLC_INTN mask = ((__CLC_INTN)1 << (24 - yexp)) - 1;
+ __CLC_INTN yodd = ((iy >> (24 - yexp)) & 0x1) != 0;
+ __CLC_INTN inty = yodd ? 1 : 2;
+ inty = (iy & mask) != 0 ? 0 : inty;
+ inty = yexp < 1 ? 0 : inty;
+ inty = yexp > 24 ? 2 : inty;
+
+ __CLC_GENTYPE signval =
+ __CLC_AS_GENTYPE((__CLC_AS_UINTN(expylogx) ^ SIGNBIT_SP32));
+ expylogx = ((inty == 1) && !xpos) ? signval : expylogx;
+ __CLC_INTN ret = __CLC_AS_INTN(expylogx);
+
+ // Corner case handling
+ __CLC_BIT_INTN y_is_ninf = iy == (__CLC_INTN)NINFBITPATT_SP32;
+ __CLC_BIT_INTN y_is_pinf = iy == (__CLC_INTN)PINFBITPATT_SP32;
+ __CLC_BIT_INTN x_is_inf = ax == (__CLC_INTN)PINFBITPATT_SP32;
+
+ ret = ax < 0x3f800000 && y_is_ninf ? PINFBITPATT_SP32 : ret;
+ ret = ax < 0x3f800000 && y_is_pinf ? 0 : ret;
+ ret = ax == 0x3f800000 && ay < PINFBITPATT_SP32 ? 0x3f800000 : ret;
+ ret = ax == 0x3f800000 && ay == PINFBITPATT_SP32 ? QNANBITPATT_SP32 : ret;
+ ret = ax > 0x3f800000 && y_is_ninf ? 0 : ret;
+ ret = ax > 0x3f800000 && y_is_pinf ? PINFBITPATT_SP32 : ret;
+ ret = ((ix < PINFBITPATT_SP32) && (ay == 0)) ? 0x3f800000 : ret;
+ ret = (x_is_inf && !ypos) ? 0 : ret;
+ ret = (x_is_inf && ypos) ? PINFBITPATT_SP32 : ret;
+ ret = (x_is_inf && y_is_pinf) ? PINFBITPATT_SP32 : ret;
+ ret = (x_is_inf && (ay == 0)) ? QNANBITPATT_SP32 : ret;
+ ret = ((ax == 0) && !ypos) ? PINFBITPATT_SP32 : ret;
+ ret = ((ax == 0) && ypos) ? 0 : ret;
+ ret = ((ax == 0) && (ay == 0)) ? QNANBITPATT_SP32 : ret;
+ ret = ((ax != 0) && !xpos) ? QNANBITPATT_SP32 : ret;
+ ret = ax > PINFBITPATT_SP32 ? ix : ret;
+ ret = ay > PINFBITPATT_SP32 ? iy : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_powr(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ const __CLC_GENTYPE real_log2_tail = 5.76999904754328540596e-08;
+ const __CLC_GENTYPE real_log2_lead = 6.93147122859954833984e-01;
+
+ __CLC_LONGN ux = __CLC_AS_LONGN(x);
+ __CLC_LONGN ax = __CLC_AS_LONGN(__clc_fabs(x));
+ __CLC_BIT_INTN xpos = ax == ux;
+
+ __CLC_LONGN uy = __CLC_AS_LONGN(y);
+ __CLC_LONGN ay = __CLC_AS_LONGN(__clc_fabs(y));
+ __CLC_BIT_INTN ypos = ay == uy;
+
+ // Extended precision log
+ __CLC_GENTYPE v, vt;
+ {
+ __CLC_INTN exp = __CLC_CONVERT_INTN(ax >> 52) - 1023;
+ __CLC_INTN mask_exp_1023 = exp == (__CLC_INTN)-1023;
+ __CLC_GENTYPE xexp = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa = ax & 0x000FFFFFFFFFFFFFL;
+
+ __CLC_LONGN temp_ux =
+ __CLC_AS_LONGN(__CLC_AS_GENTYPE(0x3ff0000000000000L | mantissa) - 1.0);
+ exp = __CLC_CONVERT_INTN((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
+ __CLC_GENTYPE xexp1 = __CLC_CONVERT_GENTYPE(exp);
+ __CLC_LONGN mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
+
+ xexp = __CLC_CONVERT_LONGN(mask_exp_1023) ? xexp1 : xexp;
+ mantissa = __CLC_CONVERT_LONGN(mask_exp_1023) ? mantissa1 : mantissa;
+
+ __CLC_LONGN rax = (mantissa & 0x000ff00000000000) +
+ ((mantissa & 0x0000080000000000) << 1);
+ __CLC_INTN index = __CLC_CONVERT_INTN(rax >> 44);
+
+ __CLC_GENTYPE F = __CLC_AS_GENTYPE(rax | 0x3FE0000000000000L);
+ __CLC_GENTYPE Y = __CLC_AS_GENTYPE(mantissa | 0x3FE0000000000000L);
+ __CLC_GENTYPE f = F - Y;
+ __CLC_GENTYPE log_h = USE_TABLE(log_f_inv_tbl_head, index);
+ __CLC_GENTYPE log_t = USE_TABLE(log_f_inv_tbl_tail, index);
+ __CLC_GENTYPE f_inv = (log_h + log_t) * f;
+ __CLC_GENTYPE r1 =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(f_inv) & 0xfffffffff8000000L);
+ __CLC_GENTYPE r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
+ __CLC_GENTYPE r = r1 + r2;
+
+ __CLC_GENTYPE poly = __clc_fma(
+ r,
+ __clc_fma(r,
+ __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
+ 1.0 / 4.0),
+ 1.0 / 3.0);
+ poly = poly * r * r * r;
+
+ __CLC_GENTYPE hr1r1 = 0.5 * r1 * r1;
+ __CLC_GENTYPE poly0h = r1 + hr1r1;
+ __CLC_GENTYPE poly0t = r1 - poly0h + hr1r1;
+ poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
+
+ log_h = USE_TABLE(powlog_tbl_head, index);
+ log_t = USE_TABLE(powlog_tbl_tail, index);
+
+ __CLC_GENTYPE resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
+ __CLC_GENTYPE resT = resT_t - poly0h;
+ __CLC_GENTYPE resH = __clc_fma(xexp, real_log2_lead, log_h);
+ __CLC_GENTYPE resT_h = poly0h;
+
+ __CLC_GENTYPE H = resT + resH;
+ __CLC_GENTYPE H_h =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(H) & 0xfffffffff8000000L);
+ __CLC_GENTYPE T =
+ (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
+ H = H_h;
+
+ __CLC_GENTYPE y_head =
+ __CLC_AS_GENTYPE(__CLC_AS_ULONGN(uy) & 0xfffffffff8000000L);
+ __CLC_GENTYPE y_tail = y - y_head;
+
+ __CLC_GENTYPE temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
+ v = __clc_fma(y_head, H, temp);
+ vt = __clc_fma(y_head, H, -v) + temp;
+ }
+
+ // Now calculate exp of (v,vt)
+
+ __CLC_GENTYPE expv;
+ {
+ const __CLC_GENTYPE max_exp_arg = 709.782712893384;
+ const __CLC_GENTYPE min_exp_arg = -745.1332191019411;
+ const __CLC_GENTYPE sixtyfour_by_lnof2 = 92.33248261689366;
+ const __CLC_GENTYPE lnof2_by_64_head = 0.010830424260348081;
+ const __CLC_GENTYPE lnof2_by_64_tail = -4.359010638708991e-10;
+
+ __CLC_GENTYPE temp = v * sixtyfour_by_lnof2;
+ __CLC_INTN n = __CLC_CONVERT_INTN(temp);
+ __CLC_GENTYPE dn = __CLC_CONVERT_GENTYPE(n);
+ __CLC_INTN j = n & 0x0000003f;
+ __CLC_INTN m = n >> 6;
+
+ __CLC_GENTYPE f1 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ __CLC_GENTYPE f2 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+ __CLC_GENTYPE f = f1 + f2;
+
+ __CLC_GENTYPE r1 = __clc_fma(dn, -lnof2_by_64_head, v);
+ __CLC_GENTYPE r2 = dn * lnof2_by_64_tail;
+ __CLC_GENTYPE r = (r1 + r2) + vt;
+
+ __CLC_GENTYPE q =
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r,
+ __clc_fma(r, 1.38889490863777199667e-03,
+ 8.33336798434219616221e-03),
+ 4.16666666662260795726e-02),
+ 1.66666666665260878863e-01),
+ 5.00000000000000008883e-01);
+ q = __clc_fma(r * r, q, r);
+
+ expv = __clc_fma(f, q, f2) + f1;
+ expv = __clc_ldexp(expv, m);
+
+ expv = v > max_exp_arg ? __CLC_AS_GENTYPE((__CLC_ULONGN)0x7FF0000000000000L)
+ : expv;
+ expv = v < min_exp_arg ? 0.0 : expv;
+ }
+
+ // See whether y is an integer.
+ // inty = 0 means not an integer.
+ // inty = 1 means odd integer.
+ // inty = 2 means even integer.
+
+ __CLC_LONGN inty;
+ {
+ __CLC_INTN yexp =
+ __CLC_CONVERT_INTN(ay >> EXPSHIFTBITS_DP64) - EXPBIAS_DP64 + 1;
+ inty = __CLC_CONVERT_LONGN(yexp < 1 ? 0 : 2);
+ inty = __CLC_CONVERT_LONGN(yexp > 53) ? 2 : inty;
+ __CLC_LONGN mask = ((__CLC_LONGN)1L << (53 - yexp)) - 1L;
+ __CLC_LONGN inty1 = (((ay & ~mask) >> (53 - yexp)) & 1L) == 1L ? 1L : 2L;
+ inty1 = (ay & mask) != 0 ? 0 : inty1;
+ inty = __CLC_CONVERT_LONGN(!(yexp < 1) && !(yexp > 53)) ? inty1 : inty;
+ }
+
+ expv *= ((inty == 1) && !xpos) ? -1.0 : 1.0;
+
+ __CLC_LONGN ret = __CLC_AS_LONGN(expv);
+
+ // Now all the edge cases
+ __CLC_BIT_INTN y_is_ninf = uy == (__CLC_LONGN)NINFBITPATT_DP64;
+ __CLC_BIT_INTN y_is_pinf = uy == (__CLC_LONGN)PINFBITPATT_DP64;
+ __CLC_BIT_INTN x_is_inf = ax == (__CLC_LONGN)PINFBITPATT_DP64;
+
+ ret = ax < 0x3ff0000000000000L && y_is_ninf ? PINFBITPATT_DP64 : ret;
+ ret = ax < 0x3ff0000000000000L && y_is_pinf ? 0L : ret;
+ ret = ax == 0x3ff0000000000000L && ay < PINFBITPATT_DP64 ? 0x3ff0000000000000L
+ : ret;
+ ret = ax == 0x3ff0000000000000L && ay == PINFBITPATT_DP64 ? QNANBITPATT_DP64
+ : ret;
+ ret = ax > 0x3ff0000000000000L && y_is_ninf ? 0L : ret;
+ ret = ax > 0x3ff0000000000000L && y_is_pinf ? PINFBITPATT_DP64 : ret;
+ ret = ux < PINFBITPATT_DP64 && ay == 0L ? 0x3ff0000000000000L : ret;
+ ret = (x_is_inf && !ypos) ? 0L : ret;
+ ret = (x_is_inf && ypos) ? PINFBITPATT_DP64 : ret;
+ ret = (x_is_inf && y_is_pinf) ? PINFBITPATT_DP64 : ret;
+ ret = ((ax == PINFBITPATT_DP64) && (ay == 0L)) ? QNANBITPATT_DP64 : ret;
+ ret = ((ax == 0L) && !ypos) ? PINFBITPATT_DP64 : ret;
+ ret = ((ax == 0L) && ypos) ? 0L : ret;
+ ret = ((ax == 0L) && (ay == 0L)) ? QNANBITPATT_DP64 : ret;
+ ret = ((ax != 0L) && !xpos) ? QNANBITPATT_DP64 : ret;
+ ret = ax > PINFBITPATT_DP64 ? ux : ret;
+ ret = ay > PINFBITPATT_DP64 ? uy : ret;
+
+ return __CLC_AS_GENTYPE(ret);
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_powr(__CLC_GENTYPE x,
+ __CLC_GENTYPE y) {
+ return __CLC_CONVERT_GENTYPE(
+ __clc_powr(__CLC_CONVERT_FLOATN(x), __CLC_CONVERT_FLOATN(y)));
+}
+
+#endif
diff --git a/libclc/clc/lib/generic/math/clc_tables.cl b/libclc/clc/lib/generic/math/clc_tables.cl
index 2d6d280f7ea06..040889e59c10e 100644
--- a/libclc/clc/lib/generic/math/clc_tables.cl
+++ b/libclc/clc/lib/generic/math/clc_tables.cl
@@ -9,6 +9,81 @@
#include <clc/internal/clc.h>
#include <clc/math/tables.h>
+DECLARE_TABLE(float, LOG_INV_TBL_EP_HEAD, 129) = {
+ 0x1.000000p+1f, 0x1.fc0000p+0f, 0x1.f80000p+0f, 0x1.f40000p+0f,
+ 0x1.f00000p+0f, 0x1.ec0000p+0f, 0x1.e80000p+0f, 0x1.e40000p+0f,
+ 0x1.e00000p+0f, 0x1.de0000p+0f, 0x1.da0000p+0f, 0x1.d60000p+0f,
+ 0x1.d40000p+0f, 0x1.d00000p+0f, 0x1.cc0000p+0f, 0x1.ca0000p+0f,
+ 0x1.c60000p+0f, 0x1.c20000p+0f, 0x1.c00000p+0f, 0x1.bc0000p+0f,
+ 0x1.ba0000p+0f, 0x1.b60000p+0f, 0x1.b40000p+0f, 0x1.b20000p+0f,
+ 0x1.ae0000p+0f, 0x1.ac0000p+0f, 0x1.a80000p+0f, 0x1.a60000p+0f,
+ 0x1.a40000p+0f, 0x1.a00000p+0f, 0x1.9e0000p+0f, 0x1.9c0000p+0f,
+ 0x1.980000p+0f, 0x1.960000p+0f, 0x1.940000p+0f, 0x1.920000p+0f,
+ 0x1.8e0000p+0f, 0x1.8c0000p+0f, 0x1.8a0000p+0f, 0x1.880000p+0f,
+ 0x1.860000p+0f, 0x1.820000p+0f, 0x1.800000p+0f, 0x1.7e0000p+0f,
+ 0x1.7c0000p+0f, 0x1.7a0000p+0f, 0x1.780000p+0f, 0x1.760000p+0f,
+ 0x1.740000p+0f, 0x1.720000p+0f, 0x1.700000p+0f, 0x1.6e0000p+0f,
+ 0x1.6c0000p+0f, 0x1.6a0000p+0f, 0x1.680000p+0f, 0x1.660000p+0f,
+ 0x1.640000p+0f, 0x1.620000p+0f, 0x1.600000p+0f, 0x1.5e0000p+0f,
+ 0x1.5c0000p+0f, 0x1.5a0000p+0f, 0x1.580000p+0f, 0x1.560000p+0f,
+ 0x1.540000p+0f, 0x1.520000p+0f, 0x1.500000p+0f, 0x1.500000p+0f,
+ 0x1.4e0000p+0f, 0x1.4c0000p+0f, 0x1.4a0000p+0f, 0x1.480000p+0f,
+ 0x1.460000p+0f, 0x1.460000p+0f, 0x1.440000p+0f, 0x1.420000p+0f,
+ 0x1.400000p+0f, 0x1.3e0000p+0f, 0x1.3e0000p+0f, 0x1.3c0000p+0f,
+ 0x1.3a0000p+0f, 0x1.380000p+0f, 0x1.380000p+0f, 0x1.360000p+0f,
+ 0x1.340000p+0f, 0x1.320000p+0f, 0x1.320000p+0f, 0x1.300000p+0f,
+ 0x1.2e0000p+0f, 0x1.2e0000p+0f, 0x1.2c0000p+0f, 0x1.2a0000p+0f,
+ 0x1.280000p+0f, 0x1.280000p+0f, 0x1.260000p+0f, 0x1.240000p+0f,
+ 0x1.240000p+0f, 0x1.220000p+0f, 0x1.200000p+0f, 0x1.200000p+0f,
+ 0x1.1e0000p+0f, 0x1.1e0000p+0f, 0x1.1c0000p+0f, 0x1.1a0000p+0f,
+ 0x1.1a0000p+0f, 0x1.180000p+0f, 0x1.180000p+0f, 0x1.160000p+0f,
+ 0x1.140000p+0f, 0x1.140000p+0f, 0x1.120000p+0f, 0x1.120000p+0f,
+ 0x1.100000p+0f, 0x1.0e0000p+0f, 0x1.0e0000p+0f, 0x1.0c0000p+0f,
+ 0x1.0c0000p+0f, 0x1.0a0000p+0f, 0x1.0a0000p+0f, 0x1.080000p+0f,
+ 0x1.080000p+0f, 0x1.060000p+0f, 0x1.060000p+0f, 0x1.040000p+0f,
+ 0x1.040000p+0f, 0x1.020000p+0f, 0x1.020000p+0f, 0x1.000000p+0f,
+ 0x1.000000p+0f,
+};
+
+CLC_TABLE_FUNCTION(float, LOG_INV_TBL_EP_HEAD, log_inv_tbl_ep_head);
+
+DECLARE_TABLE(float, LOG_INV_TBL_EP_TAIL, 129) = {
+ 0x0.000000p+0f, 0x1.fc07f0p-14f, 0x1.f81f82p-12f, 0x1.196792p-10f,
+ 0x1.f07c20p-10f, 0x1.80f660p-9f, 0x1.131ac0p-8f, 0x1.73ac90p-8f,
+ 0x1.e1e1e2p-8f, 0x1.75b8fep-10f, 0x1.cc0ed8p-9f, 0x1.7b654cp-8f,
+ 0x1.d41d42p-12f, 0x1.96b1eep-9f, 0x1.856890p-8f, 0x1.2cc158p-10f,
+ 0x1.1c71c8p-8f, 0x1.f8f01cp-8f, 0x1.c0e070p-9f, 0x1.d2b89ap-8f,
+ 0x1.9f2298p-9f, 0x1.d6c3dep-8f, 0x1.d0369ep-9f, 0x1.b20364p-15f,
+ 0x1.286bcap-8f, 0x1.5c06b2p-10f, 0x1.8ef606p-8f, 0x1.a034dap-9f,
+ 0x1.a41a42p-12f, 0x1.6d3f98p-8f, 0x1.91d2a2p-9f, 0x1.68a772p-11f,
+ 0x1.99999ap-8f, 0x1.0e4f80p-8f, 0x1.161f9ap-9f, 0x1.f693a2p-13f,
+ 0x1.9c18fap-8f, 0x1.3018d4p-8f, 0x1.9721eep-9f, 0x1.b97c2ap-10f,
+ 0x1.861862p-12f, 0x1.c977acp-8f, 0x1.818182p-8f, 0x1.405fd0p-8f,
+ 0x1.05f418p-8f, 0x1.a4411cp-9f, 0x1.499030p-9f, 0x1.f7390ep-10f,
+ 0x1.745d18p-10f, 0x1.0a1fd2p-10f, 0x1.702e06p-11f, 0x1.f76b44p-12f,
+ 0x1.6c16c2p-12f, 0x1.3cd154p-12f, 0x1.681682p-12f, 0x1.ec6a52p-12f,
+ 0x1.642c86p-11f, 0x1.fd3b80p-11f, 0x1.605816p-10f, 0x1.d6ee34p-10f,
+ 0x1.310572p-9f, 0x1.80ad60p-9f, 0x1.da4610p-9f, 0x1.1ed3c6p-8f,
+ 0x1.555556p-8f, 0x1.909490p-8f, 0x1.d07eaep-8f, 0x1.501502p-12f,
+ 0x1.7829ccp-10f, 0x1.5710e4p-9f, 0x1.fad40ap-9f, 0x1.539e3cp-8f,
+ 0x1.ae147ap-8f, 0x1.978fecp-13f, 0x1.be1958p-10f, 0x1.acc4bap-9f,
+ 0x1.414142p-8f, 0x1.b013fcp-8f, 0x1.165e72p-11f, 0x1.32b490p-9f,
+ 0x1.13b13cp-8f, 0x1.91c2c2p-8f, 0x1.381382p-12f, 0x1.31be7cp-9f,
+ 0x1.21cfb2p-8f, 0x1.ae45b6p-8f, 0x1.f1a516p-11f, 0x1.a32026p-9f,
+ 0x1.684bdap-8f, 0x1.2e025cp-15f, 0x1.3f69b0p-9f, 0x1.404ad0p-8f,
+ 0x1.e4129ep-8f, 0x1.160252p-9f, 0x1.350b88p-8f, 0x1.e22708p-8f,
+ 0x1.24924ap-9f, 0x1.45678ap-8f, 0x1.fb7812p-8f, 0x1.68e18cp-9f,
+ 0x1.7047dcp-8f, 0x1.779da0p-11f, 0x1.e0d5b4p-9f, 0x1.b4a404p-8f,
+ 0x1.ee5846p-10f, 0x1.453808p-8f, 0x1.181182p-12f, 0x1.c0d128p-9f,
+ 0x1.b1e5f8p-8f, 0x1.0be1c2p-9f, 0x1.5c8114p-8f, 0x1.ac73aep-11f,
+ 0x1.111112p-8f, 0x1.ef0110p-8f, 0x1.9ead7cp-9f, 0x1.b20a88p-8f,
+ 0x1.2e29f8p-9f, 0x1.7e6ec2p-8f, 0x1.a0429ap-10f, 0x1.53f390p-8f,
+ 0x1.084210p-10f, 0x1.3260a4p-8f, 0x1.26e978p-11f, 0x1.197f7ep-8f,
+ 0x1.041042p-12f, 0x1.091b52p-8f, 0x1.020408p-14f, 0x1.010102p-8f,
+ 0x0.000000p+0f,
+};
+CLC_TABLE_FUNCTION(float, LOG_INV_TBL_EP_TAIL, log_inv_tbl_ep_tail);
+
DECLARE_TABLE(float, LOGE_TBL_LO, 129) = {
0x0.000000p+0f, 0x1.fe0000p-8f, 0x1.fc0000p-7f, 0x1.7b8000p-6f,
0x1.f82000p-6f, 0x1.39e000p-5f, 0x1.774000p-5f, 0x1.b42000p-5f,
@@ -122,6 +197,50 @@ DECLARE_TABLE(float, LOG_INV_TBL, 129) = {
CLC_TABLE_FUNCTION(float, LOG_INV_TBL, log_inv_tbl);
+DECLARE_TABLE(float, EXP_TBL_EP_HEAD, 65) = {
+ 0x1.000000p+0f, 0x1.02c000p+0f, 0x1.058000p+0f, 0x1.084000p+0f,
+ 0x1.0b4000p+0f, 0x1.0e0000p+0f, 0x1.110000p+0f, 0x1.140000p+0f,
+ 0x1.170000p+0f, 0x1.1a0000p+0f, 0x1.1d4000p+0f, 0x1.204000p+0f,
+ 0x1.238000p+0f, 0x1.268000p+0f, 0x1.29c000p+0f, 0x1.2d0000p+0f,
+ 0x1.304000p+0f, 0x1.33c000p+0f, 0x1.370000p+0f, 0x1.3a4000p+0f,
+ 0x1.3dc000p+0f, 0x1.414000p+0f, 0x1.44c000p+0f, 0x1.484000p+0f,
+ 0x1.4bc000p+0f, 0x1.4f8000p+0f, 0x1.534000p+0f, 0x1.56c000p+0f,
+ 0x1.5a8000p+0f, 0x1.5e4000p+0f, 0x1.624000p+0f, 0x1.660000p+0f,
+ 0x1.6a0000p+0f, 0x1.6dc000p+0f, 0x1.71c000p+0f, 0x1.75c000p+0f,
+ 0x1.7a0000p+0f, 0x1.7e0000p+0f, 0x1.824000p+0f, 0x1.868000p+0f,
+ 0x1.8ac000p+0f, 0x1.8f0000p+0f, 0x1.934000p+0f, 0x1.97c000p+0f,
+ 0x1.9c4000p+0f, 0x1.a0c000p+0f, 0x1.a54000p+0f, 0x1.a9c000p+0f,
+ 0x1.ae8000p+0f, 0x1.b30000p+0f, 0x1.b7c000p+0f, 0x1.bcc000p+0f,
+ 0x1.c18000p+0f, 0x1.c64000p+0f, 0x1.cb4000p+0f, 0x1.d04000p+0f,
+ 0x1.d58000p+0f, 0x1.da8000p+0f, 0x1.dfc000p+0f, 0x1.e50000p+0f,
+ 0x1.ea4000p+0f, 0x1.ef8000p+0f, 0x1.f50000p+0f, 0x1.fa4000p+0f,
+ 0x1.000000p+1f,
+};
+
+CLC_TABLE_FUNCTION(float, EXP_TBL_EP_HEAD, exp_tbl_ep_head);
+
+DECLARE_TABLE(float, EXP_TBL_EP_TAIL, 65) = {
+ 0x0.000000p+0f, 0x1.347ceep-13f, 0x1.b0d314p-12f, 0x1.a28c3ap-11f,
+ 0x1.586cf8p-12f, 0x1.f61968p-11f, 0x1.80e808p-11f, 0x1.4d5754p-11f,
+ 0x1.5c1e3ep-11f, 0x1.adf5b6p-11f, 0x1.0e62d0p-13f, 0x1.1dc430p-11f,
+ 0x1.e9b9d4p-14f, 0x1.a2b2f0p-11f, 0x1.4efa8ep-11f, 0x1.42d372p-11f,
+ 0x1.7f0518p-11f, 0x1.164c82p-17f, 0x1.a7373ap-12f, 0x1.ed9a72p-11f,
+ 0x1.532608p-11f, 0x1.0510fap-11f, 0x1.043030p-11f, 0x1.515ae0p-11f,
+ 0x1.ed6a9ap-11f, 0x1.b2769cp-12f, 0x1.5ab4eap-15f, 0x1.a39b5ap-11f,
+ 0x1.83eea4p-11f, 0x1.b78ad6p-11f, 0x1.fac0e8p-14f, 0x1.1c412ap-11f,
+ 0x1.3cccfep-13f, 0x1.d91e32p-11f, 0x1.baf476p-11f, 0x1.f5ab20p-11f,
+ 0x1.1473eap-12f, 0x1.799b66p-11f, 0x1.89994cp-12f, 0x1.b33688p-13f,
+ 0x1.ca8454p-13f, 0x1.ae9914p-12f, 0x1.9bd866p-11f, 0x1.829fdep-12f,
+ 0x1.230546p-13f, 0x1.99ed76p-14f, 0x1.03b23ep-12f, 0x1.35aabcp-11f,
+ 0x1.3f32b4p-13f, 0x1.d15c26p-11f, 0x1.bb797cp-11f, 0x1.e904bcp-16f,
+ 0x1.9bdd84p-12f, 0x1.f8972ap-11f, 0x1.906e76p-11f, 0x1.96a502p-11f,
+ 0x1.8dcfbap-16f, 0x1.e603dap-12f, 0x1.2e66f6p-13f, 0x1.773c58p-15f,
+ 0x1.5f4548p-13f, 0x1.0df730p-11f, 0x1.d96db8p-14f, 0x1.e0c0cep-11f,
+ 0x0.000000p+0f,
+};
+
+CLC_TABLE_FUNCTION(float, EXP_TBL_EP_TAIL, exp_tbl_ep_tail);
+
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
@@ -357,4 +476,422 @@ DECLARE_TABLE(double, ATAN_JBY256_TBL_TAIL, 241) = {
CLC_TABLE_FUNCTION(double, ATAN_JBY256_TBL_TAIL, atan_jby256_tbl_tail);
+DECLARE_TABLE(double, LOG_F_INV_TBL_HEAD, 258) = {
+ 0x1.0000000000000p+1, 0x1.fe00000000000p+0, 0x1.fc00000000000p+0,
+ 0x1.fa00000000000p+0, 0x1.f800000000000p+0, 0x1.f600000000000p+0,
+ 0x1.f400000000000p+0, 0x1.f200000000000p+0, 0x1.f000000000000p+0,
+ 0x1.ee00000000000p+0, 0x1.ec00000000000p+0, 0x1.ea00000000000p+0,
+ 0x1.e900000000000p+0, 0x1.e700000000000p+0, 0x1.e500000000000p+0,
+ 0x1.e300000000000p+0, 0x1.e100000000000p+0, 0x1.e000000000000p+0,
+ 0x1.de00000000000p+0, 0x1.dc00000000000p+0, 0x1.da00000000000p+0,
+ 0x1.d900000000000p+0, 0x1.d700000000000p+0, 0x1.d500000000000p+0,
+ 0x1.d400000000000p+0, 0x1.d200000000000p+0, 0x1.d000000000000p+0,
+ 0x1.cf00000000000p+0, 0x1.cd00000000000p+0, 0x1.cb00000000000p+0,
+ 0x1.ca00000000000p+0, 0x1.c800000000000p+0, 0x1.c700000000000p+0,
+ 0x1.c500000000000p+0, 0x1.c300000000000p+0, 0x1.c200000000000p+0,
+ 0x1.c000000000000p+0, 0x1.bf00000000000p+0, 0x1.bd00000000000p+0,
+ 0x1.bc00000000000p+0, 0x1.ba00000000000p+0, 0x1.b900000000000p+0,
+ 0x1.b700000000000p+0, 0x1.b600000000000p+0, 0x1.b400000000000p+0,
+ 0x1.b300000000000p+0, 0x1.b200000000000p+0, 0x1.b000000000000p+0,
+ 0x1.af00000000000p+0, 0x1.ad00000000000p+0, 0x1.ac00000000000p+0,
+ 0x1.aa00000000000p+0, 0x1.a900000000000p+0, 0x1.a800000000000p+0,
+ 0x1.a600000000000p+0, 0x1.a500000000000p+0, 0x1.a400000000000p+0,
+ 0x1.a200000000000p+0, 0x1.a100000000000p+0, 0x1.a000000000000p+0,
+ 0x1.9e00000000000p+0, 0x1.9d00000000000p+0, 0x1.9c00000000000p+0,
+ 0x1.9a00000000000p+0, 0x1.9900000000000p+0, 0x1.9800000000000p+0,
+ 0x1.9700000000000p+0, 0x1.9500000000000p+0, 0x1.9400000000000p+0,
+ 0x1.9300000000000p+0, 0x1.9200000000000p+0, 0x1.9000000000000p+0,
+ 0x1.8f00000000000p+0, 0x1.8e00000000000p+0, 0x1.8d00000000000p+0,
+ 0x1.8b00000000000p+0, 0x1.8a00000000000p+0, 0x1.8900000000000p+0,
+ 0x1.8800000000000p+0, 0x1.8700000000000p+0, 0x1.8600000000000p+0,
+ 0x1.8400000000000p+0, 0x1.8300000000000p+0, 0x1.8200000000000p+0,
+ 0x1.8100000000000p+0, 0x1.8000000000000p+0, 0x1.7f00000000000p+0,
+ 0x1.7e00000000000p+0, 0x1.7d00000000000p+0, 0x1.7b00000000000p+0,
+ 0x1.7a00000000000p+0, 0x1.7900000000000p+0, 0x1.7800000000000p+0,
+ 0x1.7700000000000p+0, 0x1.7600000000000p+0, 0x1.7500000000000p+0,
+ 0x1.7400000000000p+0, 0x1.7300000000000p+0, 0x1.7200000000000p+0,
+ 0x1.7100000000000p+0, 0x1.7000000000000p+0, 0x1.6f00000000000p+0,
+ 0x1.6e00000000000p+0, 0x1.6d00000000000p+0, 0x1.6c00000000000p+0,
+ 0x1.6b00000000000p+0, 0x1.6a00000000000p+0, 0x1.6900000000000p+0,
+ 0x1.6800000000000p+0, 0x1.6700000000000p+0, 0x1.6600000000000p+0,
+ 0x1.6500000000000p+0, 0x1.6400000000000p+0, 0x1.6300000000000p+0,
+ 0x1.6200000000000p+0, 0x1.6100000000000p+0, 0x1.6000000000000p+0,
+ 0x1.5f00000000000p+0, 0x1.5e00000000000p+0, 0x1.5d00000000000p+0,
+ 0x1.5c00000000000p+0, 0x1.5b00000000000p+0, 0x1.5a00000000000p+0,
+ 0x1.5900000000000p+0, 0x1.5800000000000p+0, 0x1.5800000000000p+0,
+ 0x1.5700000000000p+0, 0x1.5600000000000p+0, 0x1.5500000000000p+0,
+ 0x1.5400000000000p+0, 0x1.5300000000000p+0, 0x1.5200000000000p+0,
+ 0x1.5100000000000p+0, 0x1.5000000000000p+0, 0x1.5000000000000p+0,
+ 0x1.4f00000000000p+0, 0x1.4e00000000000p+0, 0x1.4d00000000000p+0,
+ 0x1.4c00000000000p+0, 0x1.4b00000000000p+0, 0x1.4a00000000000p+0,
+ 0x1.4a00000000000p+0, 0x1.4900000000000p+0, 0x1.4800000000000p+0,
+ 0x1.4700000000000p+0, 0x1.4600000000000p+0, 0x1.4600000000000p+0,
+ 0x1.4500000000000p+0, 0x1.4400000000000p+0, 0x1.4300000000000p+0,
+ 0x1.4200000000000p+0, 0x1.4200000000000p+0, 0x1.4100000000000p+0,
+ 0x1.4000000000000p+0, 0x1.3f00000000000p+0, 0x1.3e00000000000p+0,
+ 0x1.3e00000000000p+0, 0x1.3d00000000000p+0, 0x1.3c00000000000p+0,
+ 0x1.3b00000000000p+0, 0x1.3b00000000000p+0, 0x1.3a00000000000p+0,
+ 0x1.3900000000000p+0, 0x1.3800000000000p+0, 0x1.3800000000000p+0,
+ 0x1.3700000000000p+0, 0x1.3600000000000p+0, 0x1.3500000000000p+0,
+ 0x1.3500000000000p+0, 0x1.3400000000000p+0, 0x1.3300000000000p+0,
+ 0x1.3200000000000p+0, 0x1.3200000000000p+0, 0x1.3100000000000p+0,
+ 0x1.3000000000000p+0, 0x1.3000000000000p+0, 0x1.2f00000000000p+0,
+ 0x1.2e00000000000p+0, 0x1.2e00000000000p+0, 0x1.2d00000000000p+0,
+ 0x1.2c00000000000p+0, 0x1.2b00000000000p+0, 0x1.2b00000000000p+0,
+ 0x1.2a00000000000p+0, 0x1.2900000000000p+0, 0x1.2900000000000p+0,
+ 0x1.2800000000000p+0, 0x1.2700000000000p+0, 0x1.2700000000000p+0,
+ 0x1.2600000000000p+0, 0x1.2500000000000p+0, 0x1.2500000000000p+0,
+ 0x1.2400000000000p+0, 0x1.2300000000000p+0, 0x1.2300000000000p+0,
+ 0x1.2200000000000p+0, 0x1.2100000000000p+0, 0x1.2100000000000p+0,
+ 0x1.2000000000000p+0, 0x1.2000000000000p+0, 0x1.1f00000000000p+0,
+ 0x1.1e00000000000p+0, 0x1.1e00000000000p+0, 0x1.1d00000000000p+0,
+ 0x1.1c00000000000p+0, 0x1.1c00000000000p+0, 0x1.1b00000000000p+0,
+ 0x1.1b00000000000p+0, 0x1.1a00000000000p+0, 0x1.1900000000000p+0,
+ 0x1.1900000000000p+0, 0x1.1800000000000p+0, 0x1.1800000000000p+0,
+ 0x1.1700000000000p+0, 0x1.1600000000000p+0, 0x1.1600000000000p+0,
+ 0x1.1500000000000p+0, 0x1.1500000000000p+0, 0x1.1400000000000p+0,
+ 0x1.1300000000000p+0, 0x1.1300000000000p+0, 0x1.1200000000000p+0,
+ 0x1.1200000000000p+0, 0x1.1100000000000p+0, 0x1.1100000000000p+0,
+ 0x1.1000000000000p+0, 0x1.0f00000000000p+0, 0x1.0f00000000000p+0,
+ 0x1.0e00000000000p+0, 0x1.0e00000000000p+0, 0x1.0d00000000000p+0,
+ 0x1.0d00000000000p+0, 0x1.0c00000000000p+0, 0x1.0c00000000000p+0,
+ 0x1.0b00000000000p+0, 0x1.0a00000000000p+0, 0x1.0a00000000000p+0,
+ 0x1.0900000000000p+0, 0x1.0900000000000p+0, 0x1.0800000000000p+0,
+ 0x1.0800000000000p+0, 0x1.0700000000000p+0, 0x1.0700000000000p+0,
+ 0x1.0600000000000p+0, 0x1.0600000000000p+0, 0x1.0500000000000p+0,
+ 0x1.0500000000000p+0, 0x1.0400000000000p+0, 0x1.0400000000000p+0,
+ 0x1.0300000000000p+0, 0x1.0300000000000p+0, 0x1.0200000000000p+0,
+ 0x1.0200000000000p+0, 0x1.0100000000000p+0, 0x1.0100000000000p+0,
+ 0x1.0000000000000p+0, 0x1.0000000000000p+0,
+};
+
+CLC_TABLE_FUNCTION(double, LOG_F_INV_TBL_HEAD, log_f_inv_tbl_head);
+
+DECLARE_TABLE(double, LOG_F_INV_TBL_TAIL, 258) = {
+ 0x0.0000000000000p+0, 0x1.fe01fe01fe020p-16, 0x1.fc07f01fc07f0p-14,
+ 0x1.1caa01fa11caap-12, 0x1.f81f81f81f820p-12, 0x1.8856506ddaba6p-11,
+ 0x1.196792909c560p-10, 0x1.7d9108c2ad433p-10, 0x1.f07c1f07c1f08p-10,
+ 0x1.38ff08b1c03ddp-9, 0x1.80f6603d980f6p-9, 0x1.d00f57403d5d0p-9,
+ 0x1.31abf0b7672a0p-12, 0x1.06a965d43919bp-10, 0x1.ceb240795ceb2p-10,
+ 0x1.522f3b834e67fp-9, 0x1.c3c3c3c3c3c3cp-9, 0x1.e01e01e01e01ep-12,
+ 0x1.75b8fe21a291cp-10, 0x1.403b9403b9404p-9, 0x1.cc0ed7303b5ccp-9,
+ 0x1.79118f3fc4da2p-11, 0x1.ed952e0b0ce46p-10, 0x1.95900eae56404p-9,
+ 0x1.d41d41d41d41dp-12, 0x1.cb28ff16c69aep-10, 0x1.96b1edd80e866p-9,
+ 0x1.372e225fe30d9p-11, 0x1.0ad12073615a2p-9, 0x1.cdb2c0397cdb3p-9,
+ 0x1.2cc157b864407p-10, 0x1.64cb5f7148404p-9, 0x1.c71c71c71c71cp-12,
+ 0x1.129a21a930b84p-9, 0x1.f1e0387f1e038p-9, 0x1.ad4e4ba80709bp-10,
+ 0x1.c0e070381c0e0p-9, 0x1.60fba1a362bb0p-10, 0x1.a5713280dee96p-9,
+ 0x1.3f59620f9ece9p-10, 0x1.9f22983759f23p-9, 0x1.478ac63fc8d5cp-10,
+ 0x1.ad87bb4671656p-9, 0x1.78b8efbb8148cp-10, 0x1.d0369d0369d03p-9,
+ 0x1.d212b601b3748p-10, 0x1.b2036406c80d9p-15, 0x1.29663b24547d1p-9,
+ 0x1.435e50d79435ep-11, 0x1.7d0ff2920bc03p-9, 0x1.5c06b15c06b16p-10,
+ 0x1.e3a5f0fd7f954p-9, 0x1.1dec0d4c77b03p-9, 0x1.73289870ac52ep-11,
+ 0x1.a034da034da03p-9, 0x1.d041da2292856p-10, 0x1.a41a41a41a41ap-12,
+ 0x1.8550f8a39409dp-9, 0x1.b4fe5e92c0686p-10, 0x1.a01a01a01a01ap-12,
+ 0x1.91d2a2067b23ap-9, 0x1.e7c5dada0b4e5p-10, 0x1.68a7725080ce1p-11,
+ 0x1.c49d4aa21b490p-9, 0x1.3333333333333p-9, 0x1.4bc363b03fccfp-10,
+ 0x1.c9f01970e4f81p-13, 0x1.97617c6ef5b25p-9, 0x1.161f9add3c0cap-9,
+ 0x1.319fe6cb39806p-10, 0x1.f693a1c451ab3p-13, 0x1.a9e240321a9e2p-9,
+ 0x1.3831f3831f383p-9, 0x1.949ebc4dcfc1cp-10, 0x1.80c6980c6980cp-11,
+ 0x1.f9d00c5fe7403p-9, 0x1.9721ed7e75347p-9, 0x1.381ec0313381fp-9,
+ 0x1.b97c2aec12653p-10, 0x1.09ef3024ae3bap-10, 0x1.8618618618618p-12,
+ 0x1.e0184f00c2780p-9, 0x1.92ef5657dba52p-9, 0x1.4940305494030p-9,
+ 0x1.0303030303030p-9, 0x1.8060180601806p-10, 0x1.017f405fd017fp-10,
+ 0x1.12a8ad278e8ddp-11, 0x1.7d05f417d05f4p-14, 0x1.d67245c02f7d6p-9,
+ 0x1.a4411c1d986a9p-9, 0x1.754d76c7316dfp-9, 0x1.49902f149902fp-9,
+ 0x1.21023358c1a68p-9, 0x1.f7390d2a6c406p-10, 0x1.b2b0805d5b2b1p-10,
+ 0x1.745d1745d1746p-10, 0x1.3c31507fa32c4p-10, 0x1.0a1fd1b7af017p-10,
+ 0x1.bc36ce3e0453ap-11, 0x1.702e05c0b8170p-11, 0x1.300b79300b793p-11,
+ 0x1.f76b4337c6cb1p-12, 0x1.a62681c860fb0p-12, 0x1.6c16c16c16c17p-12,
+ 0x1.490aa31a3cfc7p-12, 0x1.3cd153729043ep-12, 0x1.473a88d0bfd2ep-12,
+ 0x1.6816816816817p-12, 0x1.9f36016719f36p-12, 0x1.ec6a5122f9016p-12,
+ 0x1.27c29da5519cfp-11, 0x1.642c8590b2164p-11, 0x1.ab5c45606f00bp-11,
+ 0x1.fd3b80b11fd3cp-11, 0x1.2cda0c6ba4eaap-10, 0x1.6058160581606p-10,
+ 0x1.990d0a4b7ef87p-10, 0x1.d6ee340579d6fp-10, 0x1.0cf87d9c54a69p-9,
+ 0x1.310572620ae4cp-9, 0x1.5798c8ff522a2p-9, 0x1.80ad602b580adp-9,
+ 0x1.ac3e24799546fp-9, 0x1.da46102b1da46p-9, 0x1.5805601580560p-14,
+ 0x1.ed3c506b39a23p-12, 0x1.cbdd3e2970f60p-11, 0x1.5555555555555p-10,
+ 0x1.c979aee0bf805p-10, 0x1.21291e81fd58ep-9, 0x1.5fead500a9580p-9,
+ 0x1.a0fd5c5f02a3ap-9, 0x1.e45c223898adcp-9, 0x1.5015015015015p-12,
+ 0x1.c7b16ea64d422p-11, 0x1.7829cbc14e5e1p-10, 0x1.0877db8589720p-9,
+ 0x1.5710e4b5edceap-9, 0x1.a7dbb4d1fc1c8p-9, 0x1.fad40a57eb503p-9,
+ 0x1.3fd6bb00a5140p-11, 0x1.4e78ecb419ba9p-10, 0x1.00a44029100a4p-9,
+ 0x1.5c28f5c28f5c3p-9, 0x1.b9c68b2c0cc4ap-9, 0x1.978feb9f34381p-13,
+ 0x1.ecf163bb6500ap-11, 0x1.be1958b67ebb9p-10, 0x1.44e6157dc9a3bp-9,
+ 0x1.acc4baa3f0ddfp-9, 0x1.6a4cbcb2a247bp-13, 0x1.0505050505050p-10,
+ 0x1.e0b4439959819p-10, 0x1.6027f6027f602p-9, 0x1.d1e854b5e0db4p-9,
+ 0x1.165e7254813e2p-11, 0x1.76646a9d716efp-10, 0x1.32b48f757ce88p-9,
+ 0x1.ac1b24652a906p-9, 0x1.3b13b13b13b14p-12, 0x1.490e1eb208984p-10,
+ 0x1.2385830fec66ep-9, 0x1.a45a6cc111b7ep-9, 0x1.3813813813814p-12,
+ 0x1.56f472517b708p-10, 0x1.31be7bc0e8f2ap-9, 0x1.b9cbf3e55f044p-9,
+ 0x1.0e7d95bc609a9p-11, 0x1.9e6b3804d19e7p-10, 0x1.5c8b6af7963c2p-9,
+ 0x1.eb9dad43bf402p-9, 0x1.f1a515885fb37p-11, 0x1.0eeb1d3d76c02p-9,
+ 0x1.a320261a32026p-9, 0x1.c82ac40260390p-12, 0x1.a12f684bda12fp-10,
+ 0x1.69d43fda2962cp-9, 0x1.2e025c04b8097p-15, 0x1.42804b542804bp-10,
+ 0x1.3f69b02593f6ap-9, 0x1.df31cb46e21fap-9, 0x1.012b404ad012bp-10,
+ 0x1.23925e7820a7fp-9, 0x1.c8253c8253c82p-9, 0x1.b92ddc02526e5p-11,
+ 0x1.1602511602511p-9, 0x1.bf471439c9adfp-9, 0x1.a85c40939a85cp-11,
+ 0x1.166f9ac024d16p-9, 0x1.c44e10125e227p-9, 0x1.cebf48bbd90e5p-11,
+ 0x1.2492492492492p-9, 0x1.d6f2e2ec0b673p-9, 0x1.159e26af37c05p-10,
+ 0x1.4024540245402p-9, 0x1.f6f0243f6f024p-9, 0x1.5e60121579805p-10,
+ 0x1.68e18cf81b10fp-9, 0x1.2012012012012p-12, 0x1.c11f7047dc11fp-10,
+ 0x1.9e878ff70985ep-9, 0x1.779d9fdc3a219p-11, 0x1.1eace5c957907p-9,
+ 0x1.e0d5b450239e1p-9, 0x1.48bf073816367p-10, 0x1.694808dda5202p-9,
+ 0x1.7c67f2bae2b21p-12, 0x1.ee58469ee5847p-10, 0x1.c0233c0233c02p-9,
+ 0x1.14e02328a7012p-10, 0x1.561072057b573p-9, 0x1.1811811811812p-12,
+ 0x1.e28646f5a1060p-10, 0x1.c0d1284e6f1d7p-9, 0x1.23543f0c80459p-10,
+ 0x1.63cbeea4e1a09p-9, 0x1.b9a3fdd5c8cb8p-12, 0x1.0be1c159a76d2p-9,
+ 0x1.e1d1a688e4838p-9, 0x1.72044d72044d7p-10, 0x1.91713db81577bp-9,
+ 0x1.ac73ae9819b50p-11, 0x1.460334e904cf6p-9, 0x1.1111111111111p-12,
+ 0x1.feef80441fef0p-10, 0x1.de021fde021fep-9, 0x1.7b7eacc9686a0p-10,
+ 0x1.9ead7cd391fbcp-9, 0x1.0195609804390p-10, 0x1.641511e8d2b32p-9,
+ 0x1.222b1acf1ce96p-11, 0x1.2e29f79b47582p-9, 0x1.4f0d1682e11cdp-13,
+ 0x1.f9bb096771e4dp-10, 0x1.e5ee45dd96ae2p-9, 0x1.a0429a0429a04p-10,
+ 0x1.bb74d5f06c021p-9, 0x1.4fce404254fcep-10, 0x1.95766eacbc402p-9,
+ 0x1.0842108421084p-10, 0x1.73e5371d5c338p-9, 0x1.930523fbe3368p-11,
+ 0x1.56b38f225f6c4p-9, 0x1.26e978d4fdf3bp-11, 0x1.3dd40e4eb0cc6p-9,
+ 0x1.97f7d73404146p-12, 0x1.293982cc98af1p-9, 0x1.0410410410410p-12,
+ 0x1.18d6f048ff7e4p-9, 0x1.236a3ebc349dep-13, 0x1.0c9f8ee53d18cp-9,
+ 0x1.0204081020408p-14, 0x1.0486ca2f46ea6p-9, 0x1.0101010101010p-16,
+ 0x1.0080402010080p-9, 0x0.0000000000000p+0,
+};
+
+CLC_TABLE_FUNCTION(double, LOG_F_INV_TBL_TAIL, log_f_inv_tbl_tail);
+
+DECLARE_TABLE(double, POWLOG_TBL_HEAD, 258) = {
+ 0x0.0000000000000p+0, 0x1.ff00aa0000000p-9, 0x1.fe02a60000000p-8,
+ 0x1.7dc4750000000p-7, 0x1.fc0a8b0000000p-7, 0x1.3cea440000000p-6,
+ 0x1.7b91b00000000p-6, 0x1.b9fc020000000p-6, 0x1.f829b00000000p-6,
+ 0x1.1b0d980000000p-5, 0x1.39e87b0000000p-5, 0x1.58a5ba0000000p-5,
+ 0x1.77458f0000000p-5, 0x1.95c8300000000p-5, 0x1.b42dd70000000p-5,
+ 0x1.d276b80000000p-5, 0x1.f0a30c0000000p-5, 0x1.0759830000000p-4,
+ 0x1.16536e0000000p-4, 0x1.253f620000000p-4, 0x1.341d790000000p-4,
+ 0x1.42edcb0000000p-4, 0x1.51b0730000000p-4, 0x1.60658a0000000p-4,
+ 0x1.6f0d280000000p-4, 0x1.7da7660000000p-4, 0x1.8c345d0000000p-4,
+ 0x1.9ab4240000000p-4, 0x1.a926d30000000p-4, 0x1.b78c820000000p-4,
+ 0x1.c5e5480000000p-4, 0x1.d4313d0000000p-4, 0x1.e270760000000p-4,
+ 0x1.f0a30c0000000p-4, 0x1.fec9130000000p-4, 0x1.0671510000000p-3,
+ 0x1.0d77e70000000p-3, 0x1.1478580000000p-3, 0x1.1b72ad0000000p-3,
+ 0x1.2266f10000000p-3, 0x1.29552f0000000p-3, 0x1.303d710000000p-3,
+ 0x1.371fc20000000p-3, 0x1.3dfc2b0000000p-3, 0x1.44d2b60000000p-3,
+ 0x1.4ba36f0000000p-3, 0x1.526e5e0000000p-3, 0x1.59338d0000000p-3,
+ 0x1.5ff3070000000p-3, 0x1.66acd40000000p-3, 0x1.6d60fe0000000p-3,
+ 0x1.740f8f0000000p-3, 0x1.7ab8900000000p-3, 0x1.815c0a0000000p-3,
+ 0x1.87fa060000000p-3, 0x1.8e928d0000000p-3, 0x1.9525a90000000p-3,
+ 0x1.9bb3620000000p-3, 0x1.a23bc10000000p-3, 0x1.a8becf0000000p-3,
+ 0x1.af3c940000000p-3, 0x1.b5b5190000000p-3, 0x1.bc28670000000p-3,
+ 0x1.c296850000000p-3, 0x1.c8ff7c0000000p-3, 0x1.cf63540000000p-3,
+ 0x1.d5c2160000000p-3, 0x1.dc1bca0000000p-3, 0x1.e270760000000p-3,
+ 0x1.e8c0250000000p-3, 0x1.ef0adc0000000p-3, 0x1.f550a50000000p-3,
+ 0x1.fb91860000000p-3, 0x1.00e6c40000000p-2, 0x1.0402590000000p-2,
+ 0x1.071b850000000p-2, 0x1.0a324e0000000p-2, 0x1.0d46b50000000p-2,
+ 0x1.1058bf0000000p-2, 0x1.1368700000000p-2, 0x1.1675ca0000000p-2,
+ 0x1.1980d20000000p-2, 0x1.1c898c0000000p-2, 0x1.1f8ff90000000p-2,
+ 0x1.22941f0000000p-2, 0x1.2596010000000p-2, 0x1.2895a10000000p-2,
+ 0x1.2b93030000000p-2, 0x1.2e8e2b0000000p-2, 0x1.31871c0000000p-2,
+ 0x1.347dd90000000p-2, 0x1.3772660000000p-2, 0x1.3a64c50000000p-2,
+ 0x1.3d54fa0000000p-2, 0x1.4043080000000p-2, 0x1.432ef20000000p-2,
+ 0x1.4618bc0000000p-2, 0x1.4900680000000p-2, 0x1.4be5f90000000p-2,
+ 0x1.4ec9730000000p-2, 0x1.51aad80000000p-2, 0x1.548a2c0000000p-2,
+ 0x1.5767710000000p-2, 0x1.5a42ab0000000p-2, 0x1.5d1bdb0000000p-2,
+ 0x1.5ff3070000000p-2, 0x1.62c82f0000000p-2, 0x1.659b570000000p-2,
+ 0x1.686c810000000p-2, 0x1.6b3bb20000000p-2, 0x1.6e08ea0000000p-2,
+ 0x1.70d42e0000000p-2, 0x1.739d7f0000000p-2, 0x1.7664e10000000p-2,
+ 0x1.792a550000000p-2, 0x1.7bede00000000p-2, 0x1.7eaf830000000p-2,
+ 0x1.816f410000000p-2, 0x1.842d1d0000000p-2, 0x1.86e9190000000p-2,
+ 0x1.89a3380000000p-2, 0x1.8c5b7c0000000p-2, 0x1.8f11e80000000p-2,
+ 0x1.91c67e0000000p-2, 0x1.9479410000000p-2, 0x1.972a340000000p-2,
+ 0x1.99d9580000000p-2, 0x1.9c86b00000000p-2, 0x1.9f323e0000000p-2,
+ 0x1.a1dc060000000p-2, 0x1.a484090000000p-2, 0x1.a72a490000000p-2,
+ 0x1.a9cec90000000p-2, 0x1.ac718c0000000p-2, 0x1.af12930000000p-2,
+ 0x1.b1b1e00000000p-2, 0x1.b44f770000000p-2, 0x1.b6eb590000000p-2,
+ 0x1.b985890000000p-2, 0x1.bc1e080000000p-2, 0x1.beb4d90000000p-2,
+ 0x1.c149ff0000000p-2, 0x1.c3dd7a0000000p-2, 0x1.c66f4e0000000p-2,
+ 0x1.c8ff7c0000000p-2, 0x1.cb8e070000000p-2, 0x1.ce1af00000000p-2,
+ 0x1.d0a63a0000000p-2, 0x1.d32fe70000000p-2, 0x1.d5b7f90000000p-2,
+ 0x1.d83e720000000p-2, 0x1.dac3530000000p-2, 0x1.dd46a00000000p-2,
+ 0x1.dfc8590000000p-2, 0x1.e248810000000p-2, 0x1.e4c71a0000000p-2,
+ 0x1.e744260000000p-2, 0x1.e9bfa60000000p-2, 0x1.ec399d0000000p-2,
+ 0x1.eeb20c0000000p-2, 0x1.f128f50000000p-2, 0x1.f39e5b0000000p-2,
+ 0x1.f6123f0000000p-2, 0x1.f884a30000000p-2, 0x1.faf5880000000p-2,
+ 0x1.fd64f20000000p-2, 0x1.ffd2e00000000p-2, 0x1.011fab0000000p-1,
+ 0x1.02552a0000000p-1, 0x1.0389ee0000000p-1, 0x1.04bdf90000000p-1,
+ 0x1.05f14b0000000p-1, 0x1.0723e50000000p-1, 0x1.0855c80000000p-1,
+ 0x1.0986f40000000p-1, 0x1.0ab76b0000000p-1, 0x1.0be72e0000000p-1,
+ 0x1.0d163c0000000p-1, 0x1.0e44980000000p-1, 0x1.0f72410000000p-1,
+ 0x1.109f390000000p-1, 0x1.11cb810000000p-1, 0x1.12f7190000000p-1,
+ 0x1.1422020000000p-1, 0x1.154c3d0000000p-1, 0x1.1675ca0000000p-1,
+ 0x1.179eab0000000p-1, 0x1.18c6e00000000p-1, 0x1.19ee6b0000000p-1,
+ 0x1.1b154b0000000p-1, 0x1.1c3b810000000p-1, 0x1.1d610f0000000p-1,
+ 0x1.1e85f50000000p-1, 0x1.1faa340000000p-1, 0x1.20cdcd0000000p-1,
+ 0x1.21f0bf0000000p-1, 0x1.23130d0000000p-1, 0x1.2434b60000000p-1,
+ 0x1.2555bc0000000p-1, 0x1.2676200000000p-1, 0x1.2795e10000000p-1,
+ 0x1.28b5000000000p-1, 0x1.29d37f0000000p-1, 0x1.2af15f0000000p-1,
+ 0x1.2c0e9e0000000p-1, 0x1.2d2b400000000p-1, 0x1.2e47430000000p-1,
+ 0x1.2f62a90000000p-1, 0x1.307d730000000p-1, 0x1.3197a00000000p-1,
+ 0x1.32b1330000000p-1, 0x1.33ca2b0000000p-1, 0x1.34e2890000000p-1,
+ 0x1.35fa4e0000000p-1, 0x1.37117b0000000p-1, 0x1.38280f0000000p-1,
+ 0x1.393e0d0000000p-1, 0x1.3a53730000000p-1, 0x1.3b68440000000p-1,
+ 0x1.3c7c7f0000000p-1, 0x1.3d90260000000p-1, 0x1.3ea3390000000p-1,
+ 0x1.3fb5b80000000p-1, 0x1.40c7a40000000p-1, 0x1.41d8fe0000000p-1,
+ 0x1.42e9c60000000p-1, 0x1.43f9fe0000000p-1, 0x1.4509a50000000p-1,
+ 0x1.4618bc0000000p-1, 0x1.4727430000000p-1, 0x1.48353d0000000p-1,
+ 0x1.4942a80000000p-1, 0x1.4a4f850000000p-1, 0x1.4b5bd60000000p-1,
+ 0x1.4c679a0000000p-1, 0x1.4d72d30000000p-1, 0x1.4e7d810000000p-1,
+ 0x1.4f87a30000000p-1, 0x1.50913c0000000p-1, 0x1.519a4c0000000p-1,
+ 0x1.52a2d20000000p-1, 0x1.53aad00000000p-1, 0x1.54b2460000000p-1,
+ 0x1.55b9350000000p-1, 0x1.56bf9d0000000p-1, 0x1.57c57f0000000p-1,
+ 0x1.58cadb0000000p-1, 0x1.59cfb20000000p-1, 0x1.5ad4040000000p-1,
+ 0x1.5bd7d30000000p-1, 0x1.5cdb1d0000000p-1, 0x1.5ddde50000000p-1,
+ 0x1.5ee02a0000000p-1, 0x1.5fe1ed0000000p-1, 0x1.60e32f0000000p-1,
+ 0x1.61e3ef0000000p-1, 0x1.62e42e0000000p-1, 0x0.0000000000000p+0,
+};
+
+CLC_TABLE_FUNCTION(double, POWLOG_TBL_HEAD, powlog_tbl_head);
+
+DECLARE_TABLE(double, POWLOG_TBL_TAIL, 258) = {
+ 0x0.0000000000000p+0, 0x1.5885e0250435ap-36, 0x1.620cf11f86ed2p-33,
+ 0x1.f0214edba4a25p-32, 0x1.f807c79f3db4ep-36, 0x1.a352ba779a52bp-33,
+ 0x1.f56c46aa49fd5p-32, 0x1.ebe465fef5196p-32, 0x1.cf0660099f1f8p-31,
+ 0x1.247b2ff85945dp-30, 0x1.3fd7abf5202b6p-30, 0x1.f91c9a918d51ep-30,
+ 0x1.8cb73f118d3cap-31, 0x1.d91c7d6fad074p-30, 0x1.1971bec28d14cp-33,
+ 0x1.5b616a423c78ap-30, 0x1.162a6617cc971p-37, 0x1.66391c4c06d29p-30,
+ 0x1.d46f5c1d0c4b8p-29, 0x1.e14282df1f6d3p-29, 0x1.86f47424a660dp-30,
+ 0x1.d4c8de077753ep-29, 0x1.e0c307ed24f1cp-29, 0x1.26ea18763bdd3p-29,
+ 0x1.5cad69737c933p-29, 0x1.af62599088901p-29, 0x1.8c66c83d6b2d0p-30,
+ 0x1.880ceb36fb30fp-30, 0x1.495aac6ca17a4p-29, 0x1.761db4210878cp-29,
+ 0x1.eb78e862bac2fp-29, 0x1.9b2cd75790dd9p-30, 0x1.c55e5cbd3d50fp-29,
+ 0x1.162a6617cc971p-36, 0x1.dbeabaaa2e519p-32, 0x1.652cb7150c647p-30,
+ 0x1.9a11cb2cd2ee2p-28, 0x1.19d0ab1a28813p-29, 0x1.4bd9e80a41811p-29,
+ 0x1.214b596faa3dfp-28, 0x1.03fea46980bb8p-28, 0x1.1c8ffa5fd28c7p-28,
+ 0x1.e8f743bcd96c5p-35, 0x1.d98c5395315c6p-32, 0x1.996fa3ccfa7b2p-28,
+ 0x1.cd2af2ad13037p-30, 0x1.d0da1bd17200ep-30, 0x1.330410ba68b75p-28,
+ 0x1.4f27a790e7c41p-32, 0x1.3956a86f6ff1bp-30, 0x1.c6748723551d9p-29,
+ 0x1.500de9326cdfcp-29, 0x1.086c848df1b59p-30, 0x1.4357ead6836ffp-31,
+ 0x1.4832442408024p-29, 0x1.d10da8154b13dp-28, 0x1.9e8ad68ec8260p-28,
+ 0x1.cfbf706abaf18p-28, 0x1.fc56ac6326e23p-28, 0x1.9105e3185cf21p-28,
+ 0x1.d017fe5b19cc0p-28, 0x1.d1f6b48dd13fep-28, 0x1.0b63358a7e73ap-29,
+ 0x1.63063028c211cp-29, 0x1.e6a6886b09760p-29, 0x1.c138bb891cd03p-28,
+ 0x1.69f7722b7221ap-28, 0x1.57d8fac1a628cp-32, 0x1.c55e5cbd3d50fp-28,
+ 0x1.552d2ff48fe2ep-30, 0x1.7b8b26ca431bcp-28, 0x1.92decdc1c5f6dp-29,
+ 0x1.abc7c551aaa8cp-28, 0x1.6b540731a354bp-28, 0x1.2d341036b89efp-28,
+ 0x1.f9ab21a3a2e0fp-27, 0x1.39c871afb9fbdp-29, 0x1.e6add2c81f640p-28,
+ 0x1.35c95aa313f41p-27, 0x1.49d4582f6cc53p-29, 0x1.7574c1c07398fp-27,
+ 0x1.ba846dece9e8dp-27, 0x1.6999fafbc68e7p-30, 0x1.c9145e51b0103p-27,
+ 0x1.79ef2cb44850ap-27, 0x1.beec73de11275p-31, 0x1.ef4351af5a498p-29,
+ 0x1.5713a493b4a50p-27, 0x1.5c23a61385992p-27, 0x1.2a88309f57299p-27,
+ 0x1.530faa9ac8acep-27, 0x1.5fec2d792a758p-29, 0x1.5a517a71cbcd7p-28,
+ 0x1.707dc3e1cd9a3p-28, 0x1.a1a9f8ef43049p-28, 0x1.409d0276b3674p-27,
+ 0x1.0e2f613e85bd9p-29, 0x1.0027433001e5fp-32, 0x1.5dde2836d3265p-28,
+ 0x1.300134d7aaf04p-29, 0x1.cb7e0b42724f5p-28, 0x1.d6e93167e6308p-29,
+ 0x1.d1569b1526adbp-28, 0x1.e99fc338a1a41p-31, 0x1.eb01394a11b1cp-27,
+ 0x1.4f27a790e7c41p-31, 0x1.5ce3ca97b7af9p-29, 0x1.81f0f940ed857p-29,
+ 0x1.d36295d88857cp-27, 0x1.1aca1ec4af526p-29, 0x1.45743c7182726p-27,
+ 0x1.3c491aead337ep-29, 0x1.aef401a738931p-28, 0x1.1cede76092a29p-29,
+ 0x1.fba8f44f82bb4p-27, 0x1.46f5f7f3c3e1ap-27, 0x1.7055f86c9674bp-27,
+ 0x1.b41a92b6b6e1ap-27, 0x1.43d162e927628p-27, 0x1.466174013f9b1p-27,
+ 0x1.b05096ad69c62p-28, 0x1.0b169150faa58p-27, 0x1.cd98b1df85da7p-28,
+ 0x1.68b507b0f8fa8p-27, 0x1.8422df57499bap-27, 0x1.1351586970274p-30,
+ 0x1.17e08acba92eep-30, 0x1.6e04314dd0229p-29, 0x1.97f3097e56d1ap-27,
+ 0x1.356e655901286p-28, 0x1.cb761457f94d6p-31, 0x1.9af67a85a9dacp-28,
+ 0x1.53410931a909fp-27, 0x1.2c587206058f5p-29, 0x1.23bc358899c22p-29,
+ 0x1.d7bf8b6d223cbp-27, 0x1.7991ec5197ddbp-27, 0x1.a79e6bb3a9219p-27,
+ 0x1.a4c43ed663ec5p-28, 0x1.61b5a1484f438p-27, 0x1.b4e36f7ef0c3ap-27,
+ 0x1.15f026acd0d1bp-30, 0x1.f36b535cecf05p-28, 0x1.ffb7fbf3eb5c6p-29,
+ 0x1.e6a6886b09760p-28, 0x1.135eb27f5bbc3p-28, 0x1.70be7d6f6fa57p-27,
+ 0x1.ce43cc84ab338p-27, 0x1.c01d7aac3bd91p-27, 0x1.5c58d07961060p-27,
+ 0x1.628bcf941456ep-28, 0x1.c58b2a8461cd2p-27, 0x1.3071282fb989ap-28,
+ 0x1.20dab6a80f09cp-27, 0x1.4f8d84c397b1ep-27, 0x1.0d0ee08599e48p-27,
+ 0x1.d68787e37da36p-30, 0x1.66187d591bafcp-28, 0x1.2346600bae772p-29,
+ 0x1.90377d0d61b8ep-28, 0x1.f5e0dd966b907p-27, 0x1.9023cb79a00e2p-27,
+ 0x1.4e05158c28ad8p-27, 0x1.bfa7b08b18ae4p-28, 0x1.ef1e63db35f67p-27,
+ 0x1.ec2ae39493d4fp-31, 0x1.0afe930ab2fa0p-27, 0x1.25ff8a1810dd4p-29,
+ 0x1.69743fb1a71a5p-27, 0x1.f9cc676785571p-26, 0x1.b524da4cbf982p-26,
+ 0x1.a4c8b381535b8p-26, 0x1.839be809caf2cp-26, 0x1.0968a1cb82c13p-26,
+ 0x1.eae6a41723fb5p-26, 0x1.d9c29a380a4dbp-26, 0x1.094aa0ada625ep-27,
+ 0x1.973ad6fc108cap-26, 0x1.747322fdbab97p-27, 0x1.93692fa9d4221p-26,
+ 0x1.c5a992dfbc7d9p-26, 0x1.e1f33e102387ap-27, 0x1.64fbef14c048cp-27,
+ 0x1.490f513ca5e3bp-27, 0x1.7a6af4d4c799dp-28, 0x1.7574c1c07398fp-26,
+ 0x1.7b133417f8c1cp-26, 0x1.feb9e0c176514p-26, 0x1.19f25bb3172f7p-27,
+ 0x1.5f68a7bbfb852p-27, 0x1.ee278497929f1p-26, 0x1.ccee006109d58p-26,
+ 0x1.ce081a07bd8b3p-26, 0x1.70e12981817b8p-26, 0x1.92ab6d93503d0p-29,
+ 0x1.8cb7dd7c3b61ep-26, 0x1.efafd0a0b78dap-27, 0x1.e907267c4288ep-26,
+ 0x1.d31ef96780875p-26, 0x1.3430dfcd2ad50p-29, 0x1.44d88d75bc1f9p-28,
+ 0x1.bec0f055e04fcp-26, 0x1.d85611590b9adp-26, 0x1.320568e583229p-32,
+ 0x1.a891d1772f538p-26, 0x1.2edc9dabba74dp-29, 0x1.b9009a1015086p-27,
+ 0x1.2a12a8c5b1a19p-26, 0x1.a7885f0fdac85p-28, 0x1.f4ffcd43ac691p-26,
+ 0x1.2243ae2640aadp-26, 0x1.46513299035d3p-26, 0x1.b39c3a62dd725p-26,
+ 0x1.ba6dd40049f51p-26, 0x1.51d1ed7177409p-27, 0x1.cb0f2fd7f5216p-26,
+ 0x1.ab150cd4e2213p-28, 0x1.cfd7bf3193844p-26, 0x1.3fff8455f1dbdp-26,
+ 0x1.fee640b905fc9p-26, 0x1.4e2adf548084cp-26, 0x1.b597adc1ecdd2p-28,
+ 0x1.345bd096d3a75p-27, 0x1.101b9d2453c8bp-26, 0x1.08ce55cc8c979p-26,
+ 0x1.bbf017e595f71p-26, 0x1.7ce733bd393dcp-28, 0x1.33bb0a503f8a1p-29,
+ 0x1.0e2f613e85bd9p-28, 0x1.e67555a635b3cp-26, 0x1.ea88df73d5e8bp-29,
+ 0x1.d17e03bda18a8p-28, 0x1.b607d76044f7ep-26, 0x1.2adc4e71bc2fcp-26,
+ 0x1.f99dc7362d1d9p-26, 0x1.473fa008e6a6ap-26, 0x1.b75bb09cb0985p-29,
+ 0x1.ea04dd10b9abap-26, 0x1.802d0d6979674p-26, 0x1.74688ccd99094p-30,
+ 0x1.96f16abb9df22p-27, 0x1.6e66df2aa374fp-27, 0x1.e66525ea4550ap-27,
+ 0x1.2d02f34f20cbdp-27, 0x1.6cfce65047188p-27, 0x1.9b78c842d58b8p-28,
+ 0x1.735e624c24bc9p-27, 0x1.7eba1f7dd1adfp-27, 0x1.86b3e59f65355p-26,
+ 0x1.ce38e637f1b4dp-30, 0x1.8d82ec919edc7p-26, 0x1.c52648ddcfa37p-27,
+ 0x1.2482ceae1ac12p-26, 0x1.5a312311aba4fp-26, 0x1.11e236329f225p-27,
+ 0x1.b48c8cd2f246cp-26, 0x1.efa39ef35793cp-25, 0x0.0000000000000p+0,
+};
+
+CLC_TABLE_FUNCTION(double, POWLOG_TBL_TAIL, powlog_tbl_tail);
+
+DECLARE_TABLE(double, TWO_TO_JBY64_EP_HEAD, 64) = {
+ 0x1.0000000000000p+0, 0x1.02c9a30000000p+0, 0x1.059b0d0000000p+0,
+ 0x1.0874510000000p+0, 0x1.0b55860000000p+0, 0x1.0e3ec30000000p+0,
+ 0x1.11301d0000000p+0, 0x1.1429aa0000000p+0, 0x1.172b830000000p+0,
+ 0x1.1a35be0000000p+0, 0x1.1d48730000000p+0, 0x1.2063b80000000p+0,
+ 0x1.2387a60000000p+0, 0x1.26b4560000000p+0, 0x1.29e9df0000000p+0,
+ 0x1.2d285a0000000p+0, 0x1.306fe00000000p+0, 0x1.33c08b0000000p+0,
+ 0x1.371a730000000p+0, 0x1.3a7db30000000p+0, 0x1.3dea640000000p+0,
+ 0x1.4160a20000000p+0, 0x1.44e0860000000p+0, 0x1.486a2b0000000p+0,
+ 0x1.4bfdad0000000p+0, 0x1.4f9b270000000p+0, 0x1.5342b50000000p+0,
+ 0x1.56f4730000000p+0, 0x1.5ab07d0000000p+0, 0x1.5e76f10000000p+0,
+ 0x1.6247eb0000000p+0, 0x1.6623880000000p+0, 0x1.6a09e60000000p+0,
+ 0x1.6dfb230000000p+0, 0x1.71f75e0000000p+0, 0x1.75feb50000000p+0,
+ 0x1.7a11470000000p+0, 0x1.7e2f330000000p+0, 0x1.8258990000000p+0,
+ 0x1.868d990000000p+0, 0x1.8ace540000000p+0, 0x1.8f1ae90000000p+0,
+ 0x1.93737b0000000p+0, 0x1.97d8290000000p+0, 0x1.9c49180000000p+0,
+ 0x1.a0c6670000000p+0, 0x1.a5503b0000000p+0, 0x1.a9e6b50000000p+0,
+ 0x1.ae89f90000000p+0, 0x1.b33a2b0000000p+0, 0x1.b7f76f0000000p+0,
+ 0x1.bcc1e90000000p+0, 0x1.c199bd0000000p+0, 0x1.c67f120000000p+0,
+ 0x1.cb720d0000000p+0, 0x1.d072d40000000p+0, 0x1.d5818d0000000p+0,
+ 0x1.da9e600000000p+0, 0x1.dfc9730000000p+0, 0x1.e502ee0000000p+0,
+ 0x1.ea4afa0000000p+0, 0x1.efa1be0000000p+0, 0x1.f507650000000p+0,
+ 0x1.fa7c180000000p+0,
+};
+
+CLC_TABLE_FUNCTION(double, TWO_TO_JBY64_EP_HEAD, two_to_jby64_ep_tbl_head);
+
+DECLARE_TABLE(double, TWO_TO_JBY64_EP_TAIL, 64) = {
+ 0x0.0000000000000p+0, 0x1.cef00c1dcdef9p-25, 0x1.8ac2ba1d73e2ap-27,
+ 0x1.0eb37901186bep-25, 0x1.9f3121ec53172p-25, 0x1.69e8d10103a17p-27,
+ 0x1.25b50a4ebbf1ap-32, 0x1.d525bbf668203p-25, 0x1.8faa2f5b9bef9p-25,
+ 0x1.6df96ea796d31p-25, 0x1.68b9aa7805b80p-28, 0x1.0c519ac771dd6p-25,
+ 0x1.ceac470cd83f5p-25, 0x1.789f37495e99cp-26, 0x1.47f7b84b09745p-26,
+ 0x1.b900c2d002475p-26, 0x1.4636e2a5bd1abp-25, 0x1.320b7fa64e430p-27,
+ 0x1.ceaa72a9c5154p-26, 0x1.3967fdba86f24p-26, 0x1.82468446b6824p-25,
+ 0x1.f72e29f84325bp-28, 0x1.8624b40c4dbd0p-30, 0x1.704f3404f068ep-26,
+ 0x1.4d8a89c750e5ep-26, 0x1.a74b29ab4cf62p-26, 0x1.a753e077c2a0fp-26,
+ 0x1.ad49f699bb2c0p-26, 0x1.a90a852b19260p-25, 0x1.6b48521ba6f93p-26,
+ 0x1.d2ac258f87d03p-31, 0x1.2a91124893ecfp-27, 0x1.9fcef32422cbep-26,
+ 0x1.8ca345de441c5p-25, 0x1.1d8bee7ba46e1p-25, 0x1.9099f22fdba6ap-26,
+ 0x1.f580c36bea881p-27, 0x1.b3d398841740ap-26, 0x1.2999c25159f11p-25,
+ 0x1.68925d901c83bp-25, 0x1.15506dadd3e2ap-27, 0x1.22aee6c57304ep-25,
+ 0x1.9b8bc9e8a0387p-29, 0x1.fbc9c9f173d24p-25, 0x1.51f8480e3e235p-27,
+ 0x1.6bbcac96535b5p-25, 0x1.1f12ae45a1224p-27, 0x1.5e7f6fd0fac90p-26,
+ 0x1.2b5a75abd0e69p-25, 0x1.09e2bf5ed7fa1p-25, 0x1.7daf237553d84p-27,
+ 0x1.2f074891ee83dp-30, 0x1.b0aa538444196p-25, 0x1.cafa29694426fp-25,
+ 0x1.9df20d22a0797p-25, 0x1.40f12f71a1e45p-25, 0x1.9f7490e4bb40bp-25,
+ 0x1.ed9942b84600dp-27, 0x1.bdcdaf5cb4656p-27, 0x1.e2cffd89cf44cp-26,
+ 0x1.52486cc2c7b9dp-27, 0x1.cc2b44eee3fa4p-25, 0x1.6dc8a80ce9f09p-25,
+ 0x1.9e90d82e90a7ep-28,
+};
+
+CLC_TABLE_FUNCTION(double, TWO_TO_JBY64_EP_TAIL, two_to_jby64_ep_tbl_tail);
+
#endif // cl_khr_fp64
diff --git a/libclc/clspv/lib/SOURCES b/libclc/clspv/lib/SOURCES
index 4354cb6f4b32c..c0a84f2a554f0 100644
--- a/libclc/clspv/lib/SOURCES
+++ b/libclc/clspv/lib/SOURCES
@@ -17,9 +17,6 @@ subnormal_config.cl
../../generic/lib/math/atanpi.cl
../../generic/lib/math/cbrt.cl
../../generic/lib/math/clc_exp10.cl
-../../generic/lib/math/clc_pow.cl
-../../generic/lib/math/clc_pown.cl
-../../generic/lib/math/clc_powr.cl
../../generic/lib/math/clc_rootn.cl
../../generic/lib/math/clc_tan.cl
../../generic/lib/math/clc_tanpi.cl
diff --git a/libclc/generic/include/clc/math/pown.h b/libclc/generic/include/clc/math/pown.h
index 42f5dd4b4301a..a448f693c9096 100644
--- a/libclc/generic/include/clc/math/pown.h
+++ b/libclc/generic/include/clc/math/pown.h
@@ -6,6 +6,10 @@
//
//===----------------------------------------------------------------------===//
-#define __CLC_BODY <clc/math/pown.inc>
+#define __CLC_FUNCTION pown
+#define __CLC_BODY <clc/math/binary_decl_with_int_second_arg.inc>
+
#include <clc/math/gentype.inc>
+
#undef __CLC_BODY
+#undef __CLC_FUNCTION
diff --git a/libclc/generic/include/math/clc_pown.inc b/libclc/generic/include/math/clc_pown.inc
deleted file mode 100644
index cf90fe7ec9f4b..0000000000000
--- a/libclc/generic/include/math/clc_pown.inc
+++ /dev/null
@@ -1,9 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __clc_pown(__CLC_GENTYPE a, __CLC_INTN b);
diff --git a/libclc/generic/lib/SOURCES b/libclc/generic/lib/SOURCES
index 896c25b43078a..1c3aff16035c5 100644
--- a/libclc/generic/lib/SOURCES
+++ b/libclc/generic/lib/SOURCES
@@ -156,11 +156,8 @@ math/native_sqrt.cl
math/native_tan.cl
math/tables.cl
math/nextafter.cl
-math/clc_pow.cl
math/pow.cl
-math/clc_pown.cl
math/pown.cl
-math/clc_powr.cl
math/powr.cl
math/remainder.cl
math/remquo.cl
diff --git a/libclc/generic/lib/math/clc_exp10.cl b/libclc/generic/lib/math/clc_exp10.cl
index 9d70efe990216..0b6f2f9f26db1 100644
--- a/libclc/generic/lib/math/clc_exp10.cl
+++ b/libclc/generic/lib/math/clc_exp10.cl
@@ -127,8 +127,9 @@ _CLC_DEF _CLC_OVERLOAD double __clc_exp10(double x) {
0x1.0000000000000p-1),
1.0);
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- z2 = __clc_fma(tv.s0 + tv.s1, z2, tv.s1) + tv.s0;
+ double tv0 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ double tv1 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+ z2 = __clc_fma(tv0 + tv1, z2, tv1) + tv0;
int small_value = (m < -1022) || ((m == -1022) && (z2 < 1.0));
diff --git a/libclc/generic/lib/math/clc_pow.cl b/libclc/generic/lib/math/clc_pow.cl
deleted file mode 100644
index 9205133bd7a24..0000000000000
--- a/libclc/generic/lib/math/clc_pow.cl
+++ /dev/null
@@ -1,410 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-#include <clc/clc.h>
-#include <clc/clc_convert.h>
-#include <clc/clcmacro.h>
-#include <clc/math/clc_fabs.h>
-#include <clc/math/clc_fma.h>
-#include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
-#include <clc/math/math.h>
-#include <clc/math/tables.h>
-
-/*
- compute pow using log and exp
- x^y = exp(y * log(x))
-
- we take care not to lose precision in the intermediate steps
-
- When computing log, calculate it in splits,
-
- r = f * (p_invead + p_inv_tail)
- r = rh + rt
-
- calculate log polynomial using r, in end addition, do
- poly = poly + ((rh-r) + rt)
-
- lth = -r
- ltt = ((xexp * log2_t) - poly) + logT
- lt = lth + ltt
-
- lh = (xexp * log2_h) + logH
- l = lh + lt
-
- Calculate final log answer as gh and gt,
- gh = l & higher-half bits
- gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
-
- yh = y & higher-half bits
- yt = y - yh
-
- Before entering computation of exp,
- vs = ((yt*gt + yt*gh) + yh*gt)
- v = vs + yh*gh
- vt = ((yh*gh - v) + vs)
-
- In calculation of exp, add vt to r that is used for poly
- At the end of exp, do
- ((((expT * poly) + expT) + expH*poly) + expH)
-*/
-
-_CLC_DEF _CLC_OVERLOAD float __clc_pow(float x, float y) {
-
- int ix = __clc_as_int(x);
- int ax = ix & EXSIGNBIT_SP32;
- int xpos = ix == ax;
-
- int iy = __clc_as_int(y);
- int ay = iy & EXSIGNBIT_SP32;
- int ypos = iy == ay;
-
- /* Extra precise log calculation
- * First handle case that x is close to 1
- */
- float r = 1.0f - __clc_as_float(ax);
- int near1 = __clc_fabs(r) < 0x1.0p-4f;
- float r2 = r * r;
-
- /* Coefficients are just 1/3, 1/4, 1/5 and 1/6 */
- float poly = __clc_mad(
- r,
- __clc_mad(r,
- __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
- 0x1.99999ap-3f),
- 0x1.000000p-2f),
- 0x1.555556p-2f);
-
- poly *= r2 * r;
-
- float lth_near1 = -r2 * 0.5f;
- float ltt_near1 = -poly;
- float lt_near1 = lth_near1 + ltt_near1;
- float lh_near1 = -r;
- float l_near1 = lh_near1 + lt_near1;
-
- /* Computations for x not near 1 */
- int m = (int)(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
- float mf = (float)m;
- int ixs = __clc_as_int(__clc_as_float(ax | 0x3f800000) - 1.0f);
- float mfs = (float)((ixs >> EXPSHIFTBITS_SP32) - 253);
- int c = m == -127;
- int ixn = c ? ixs : ax;
- float mfn = c ? mfs : mf;
-
- int indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
-
- /* F - Y */
- float f = __clc_as_float(0x3f000000 | indx) -
- __clc_as_float(0x3f000000 | (ixn & MANTBITS_SP32));
-
- indx = indx >> 16;
- float2 tv = USE_TABLE(log_inv_tbl_ep, indx);
- float rh = f * tv.s0;
- float rt = f * tv.s1;
- r = rh + rt;
-
- poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
- (r * r);
- poly += (rh - r) + rt;
-
- const float LOG2_HEAD = 0x1.62e000p-1f; /* 0.693115234 */
- const float LOG2_TAIL = 0x1.0bfbe8p-15f; /* 0.0000319461833 */
- tv.s0 = USE_TABLE(loge_tbl_lo, indx);
- tv.s1 = USE_TABLE(loge_tbl_hi, indx);
- float lth = -r;
- float ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + tv.s1;
- float lt = lth + ltt;
- float lh = __clc_mad(mfn, LOG2_HEAD, tv.s0);
- float l = lh + lt;
-
- /* Select near 1 or not */
- lth = near1 ? lth_near1 : lth;
- ltt = near1 ? ltt_near1 : ltt;
- lt = near1 ? lt_near1 : lt;
- lh = near1 ? lh_near1 : lh;
- l = near1 ? l_near1 : l;
-
- float gh = __clc_as_float(__clc_as_int(l) & 0xfffff000);
- float gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
-
- float yh = __clc_as_float(iy & 0xfffff000);
-
- float yt = y - yh;
-
- float ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
- float ylogx = __clc_mad(yh, gh, ylogx_s);
- float ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
-
- /* Extra precise exp of ylogx */
- /* 64/log2 : 92.332482616893657 */
- const float R_64_BY_LOG2 = 0x1.715476p+6f;
- int n = __clc_convert_int(ylogx * R_64_BY_LOG2);
- float nf = (float)n;
-
- int j = n & 0x3f;
- m = n >> 6;
- int m2 = m << EXPSHIFTBITS_SP32;
-
- /* log2/64 lead: 0.0108032227 */
- const float R_LOG2_BY_64_LD = 0x1.620000p-7f;
- /* log2/64 tail: 0.0000272020388 */
- const float R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
- r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
- ylogx_t;
-
- /* Truncated Taylor series for e^r */
- poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
- 0x1.000000p-1f),
- r * r, r);
-
- tv = USE_TABLE(exp_tbl_ep, j);
-
- float expylogx =
- __clc_mad(tv.s0, poly, __clc_mad(tv.s1, poly, tv.s1)) + tv.s0;
- float sexpylogx = expylogx * __clc_as_float(0x1 << (m + 149));
- float texpylogx = __clc_as_float(__clc_as_int(expylogx) + m2);
- expylogx = m < -125 ? sexpylogx : texpylogx;
-
- /* Result is +-Inf if (ylogx + ylogx_t) > 128*log2 */
- expylogx = (ylogx > 0x1.62e430p+6f) |
- (ylogx == 0x1.62e430p+6f & ylogx_t > -0x1.05c610p-22f)
- ? __clc_as_float(PINFBITPATT_SP32)
- : expylogx;
-
- /* Result is 0 if ylogx < -149*log2 */
- expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
-
- /* Classify y:
- * inty = 0 means not an integer.
- * inty = 1 means odd integer.
- * inty = 2 means even integer.
- */
-
- int yexp = (int)(ay >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32 + 1;
- int mask = (1 << (24 - yexp)) - 1;
- int yodd = ((iy >> (24 - yexp)) & 0x1) != 0;
- int inty = yodd ? 1 : 2;
- inty = (iy & mask) != 0 ? 0 : inty;
- inty = yexp < 1 ? 0 : inty;
- inty = yexp > 24 ? 2 : inty;
-
- float signval = __clc_as_float((__clc_as_uint(expylogx) ^ SIGNBIT_SP32));
- expylogx = ((inty == 1) & !xpos) ? signval : expylogx;
- int ret = __clc_as_int(expylogx);
-
- /* Corner case handling */
- ret = (!xpos & (inty == 0)) ? QNANBITPATT_SP32 : ret;
- ret = ax < 0x3f800000 & iy == NINFBITPATT_SP32 ? PINFBITPATT_SP32 : ret;
- ret = ax > 0x3f800000 & iy == NINFBITPATT_SP32 ? 0 : ret;
- ret = ax < 0x3f800000 & iy == PINFBITPATT_SP32 ? 0 : ret;
- ret = ax > 0x3f800000 & iy == PINFBITPATT_SP32 ? PINFBITPATT_SP32 : ret;
- int xinf = xpos ? PINFBITPATT_SP32 : NINFBITPATT_SP32;
- ret = ((ax == 0) & !ypos & (inty == 1)) ? xinf : ret;
- ret = ((ax == 0) & !ypos & (inty != 1)) ? PINFBITPATT_SP32 : ret;
- int xzero = xpos ? 0 : 0x80000000;
- ret = ((ax == 0) & ypos & (inty == 1)) ? xzero : ret;
- ret = ((ax == 0) & ypos & (inty != 1)) ? 0 : ret;
- ret = ((ax == 0) & (iy == NINFBITPATT_SP32)) ? PINFBITPATT_SP32 : ret;
- ret = ((ix == 0xbf800000) & (ay == PINFBITPATT_SP32)) ? 0x3f800000 : ret;
- ret = ((ix == NINFBITPATT_SP32) & !ypos & (inty == 1)) ? 0x80000000 : ret;
- ret = ((ix == NINFBITPATT_SP32) & !ypos & (inty != 1)) ? 0 : ret;
- ret =
- ((ix == NINFBITPATT_SP32) & ypos & (inty == 1)) ? NINFBITPATT_SP32 : ret;
- ret =
- ((ix == NINFBITPATT_SP32) & ypos & (inty != 1)) ? PINFBITPATT_SP32 : ret;
- ret = ((ix == PINFBITPATT_SP32) & !ypos) ? 0 : ret;
- ret = ((ix == PINFBITPATT_SP32) & ypos) ? PINFBITPATT_SP32 : ret;
- ret = (ax > PINFBITPATT_SP32) ? ix : ret;
- ret = (ay > PINFBITPATT_SP32) ? iy : ret;
- ret = ay == 0 ? 0x3f800000 : ret;
- ret = ix == 0x3f800000 ? 0x3f800000 : ret;
-
- return __clc_as_float(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, float, __clc_pow, float, float)
-
-#ifdef cl_khr_fp64
-_CLC_DEF _CLC_OVERLOAD double __clc_pow(double x, double y) {
- const double real_log2_tail = 5.76999904754328540596e-08;
- const double real_log2_lead = 6.93147122859954833984e-01;
-
- long ux = __clc_as_long(x);
- long ax = ux & (~SIGNBIT_DP64);
- int xpos = ax == ux;
-
- long uy = __clc_as_long(y);
- long ay = uy & (~SIGNBIT_DP64);
- int ypos = ay == uy;
-
- // Extended precision log
- double v, vt;
- {
- int exp = (int)(ax >> 52) - 1023;
- int mask_exp_1023 = exp == -1023;
- double xexp = (double)exp;
- long mantissa = ax & 0x000FFFFFFFFFFFFFL;
-
- long temp_ux =
- __clc_as_long(__clc_as_double(0x3ff0000000000000L | mantissa) - 1.0);
- exp = ((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
- double xexp1 = (double)exp;
- long mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
-
- xexp = mask_exp_1023 ? xexp1 : xexp;
- mantissa = mask_exp_1023 ? mantissa1 : mantissa;
-
- long rax = (mantissa & 0x000ff00000000000) +
- ((mantissa & 0x0000080000000000) << 1);
- int index = rax >> 44;
-
- double F = __clc_as_double(rax | 0x3FE0000000000000L);
- double Y = __clc_as_double(mantissa | 0x3FE0000000000000L);
- double f = F - Y;
- double2 tv = USE_TABLE(log_f_inv_tbl, index);
- double log_h = tv.s0;
- double log_t = tv.s1;
- double f_inv = (log_h + log_t) * f;
- double r1 = __clc_as_double(__clc_as_long(f_inv) & 0xfffffffff8000000L);
- double r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
- double r = r1 + r2;
-
- double poly = __clc_fma(
- r,
- __clc_fma(r,
- __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
- 1.0 / 4.0),
- 1.0 / 3.0);
- poly = poly * r * r * r;
-
- double hr1r1 = 0.5 * r1 * r1;
- double poly0h = r1 + hr1r1;
- double poly0t = r1 - poly0h + hr1r1;
- poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
-
- tv = USE_TABLE(powlog_tbl, index);
- log_h = tv.s0;
- log_t = tv.s1;
-
- double resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
- double resT = resT_t - poly0h;
- double resH = __clc_fma(xexp, real_log2_lead, log_h);
- double resT_h = poly0h;
-
- double H = resT + resH;
- double H_h = __clc_as_double(__clc_as_long(H) & 0xfffffffff8000000L);
- double T = (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
- H = H_h;
-
- double y_head = __clc_as_double(uy & 0xfffffffff8000000L);
- double y_tail = y - y_head;
-
- double temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
- v = __clc_fma(y_head, H, temp);
- vt = __clc_fma(y_head, H, -v) + temp;
- }
-
- // Now calculate exp of (v,vt)
-
- double expv;
- {
- const double max_exp_arg = 709.782712893384;
- const double min_exp_arg = -745.1332191019411;
- const double sixtyfour_by_lnof2 = 92.33248261689366;
- const double lnof2_by_64_head = 0.010830424260348081;
- const double lnof2_by_64_tail = -4.359010638708991e-10;
-
- double temp = v * sixtyfour_by_lnof2;
- int n = (int)temp;
- double dn = (double)n;
- int j = n & 0x0000003f;
- int m = n >> 6;
-
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- double f1 = tv.s0;
- double f2 = tv.s1;
- double f = f1 + f2;
-
- double r1 = __clc_fma(dn, -lnof2_by_64_head, v);
- double r2 = dn * lnof2_by_64_tail;
- double r = (r1 + r2) + vt;
-
- double q =
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r, 1.38889490863777199667e-03,
- 8.33336798434219616221e-03),
- 4.16666666662260795726e-02),
- 1.66666666665260878863e-01),
- 5.00000000000000008883e-01);
- q = __clc_fma(r * r, q, r);
-
- expv = __clc_fma(f, q, f2) + f1;
- expv = ldexp(expv, m);
-
- expv = v > max_exp_arg ? __clc_as_double(0x7FF0000000000000L) : expv;
- expv = v < min_exp_arg ? 0.0 : expv;
- }
-
- // See whether y is an integer.
- // inty = 0 means not an integer.
- // inty = 1 means odd integer.
- // inty = 2 means even integer.
-
- int inty;
- {
- int yexp = (int)(ay >> EXPSHIFTBITS_DP64) - EXPBIAS_DP64 + 1;
- inty = yexp < 1 ? 0 : 2;
- inty = yexp > 53 ? 2 : inty;
- long mask = (1L << (53 - yexp)) - 1L;
- int inty1 = (((ay & ~mask) >> (53 - yexp)) & 1L) == 1L ? 1 : 2;
- inty1 = (ay & mask) != 0 ? 0 : inty1;
- inty = !(yexp < 1) & !(yexp > 53) ? inty1 : inty;
- }
-
- expv *= (inty == 1) & !xpos ? -1.0 : 1.0;
-
- long ret = __clc_as_long(expv);
-
- // Now all the edge cases
- ret = !xpos & (inty == 0) ? QNANBITPATT_DP64 : ret;
- ret = ax < 0x3ff0000000000000L & uy == NINFBITPATT_DP64 ? PINFBITPATT_DP64
- : ret;
- ret = ax > 0x3ff0000000000000L & uy == NINFBITPATT_DP64 ? 0L : ret;
- ret = ax < 0x3ff0000000000000L & uy == PINFBITPATT_DP64 ? 0L : ret;
- ret = ax > 0x3ff0000000000000L & uy == PINFBITPATT_DP64 ? PINFBITPATT_DP64
- : ret;
- long xinf = xpos ? PINFBITPATT_DP64 : NINFBITPATT_DP64;
- ret = ((ax == 0L) & !ypos & (inty == 1)) ? xinf : ret;
- ret = ((ax == 0L) & !ypos & (inty != 1)) ? PINFBITPATT_DP64 : ret;
- long xzero = xpos ? 0L : 0x8000000000000000L;
- ret = ((ax == 0L) & ypos & (inty == 1)) ? xzero : ret;
- ret = ((ax == 0L) & ypos & (inty != 1)) ? 0L : ret;
- ret = ((ax == 0L) & (uy == NINFBITPATT_DP64)) ? PINFBITPATT_DP64 : ret;
- ret = ((ux == 0xbff0000000000000L) & (ay == PINFBITPATT_DP64))
- ? 0x3ff0000000000000L
- : ret;
- ret = ((ux == NINFBITPATT_DP64) & !ypos & (inty == 1)) ? 0x8000000000000000L
- : ret;
- ret = ((ux == NINFBITPATT_DP64) & !ypos & (inty != 1)) ? 0L : ret;
- ret =
- ((ux == NINFBITPATT_DP64) & ypos & (inty == 1)) ? NINFBITPATT_DP64 : ret;
- ret =
- ((ux == NINFBITPATT_DP64) & ypos & (inty != 1)) ? PINFBITPATT_DP64 : ret;
- ret = (ux == PINFBITPATT_DP64) & !ypos ? 0L : ret;
- ret = (ux == PINFBITPATT_DP64) & ypos ? PINFBITPATT_DP64 : ret;
- ret = ax > PINFBITPATT_DP64 ? ux : ret;
- ret = ay > PINFBITPATT_DP64 ? uy : ret;
- ret = ay == 0L ? 0x3ff0000000000000L : ret;
- ret = ux == 0x3ff0000000000000L ? 0x3ff0000000000000L : ret;
-
- return __clc_as_double(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, double, __clc_pow, double, double)
-#endif
diff --git a/libclc/generic/lib/math/clc_pown.cl b/libclc/generic/lib/math/clc_pown.cl
deleted file mode 100644
index e6ea645121f87..0000000000000
--- a/libclc/generic/lib/math/clc_pown.cl
+++ /dev/null
@@ -1,390 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-#include <clc/clc.h>
-#include <clc/clc_convert.h>
-#include <clc/clcmacro.h>
-#include <clc/math/clc_fabs.h>
-#include <clc/math/clc_fma.h>
-#include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
-#include <clc/math/math.h>
-#include <clc/math/tables.h>
-
-// compute pow using log and exp
-// x^y = exp(y * log(x))
-//
-// we take care not to lose precision in the intermediate steps
-//
-// When computing log, calculate it in splits,
-//
-// r = f * (p_invead + p_inv_tail)
-// r = rh + rt
-//
-// calculate log polynomial using r, in end addition, do
-// poly = poly + ((rh-r) + rt)
-//
-// lth = -r
-// ltt = ((xexp * log2_t) - poly) + logT
-// lt = lth + ltt
-//
-// lh = (xexp * log2_h) + logH
-// l = lh + lt
-//
-// Calculate final log answer as gh and gt,
-// gh = l & higher-half bits
-// gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
-//
-// yh = y & higher-half bits
-// yt = y - yh
-//
-// Before entering computation of exp,
-// vs = ((yt*gt + yt*gh) + yh*gt)
-// v = vs + yh*gh
-// vt = ((yh*gh - v) + vs)
-//
-// In calculation of exp, add vt to r that is used for poly
-// At the end of exp, do
-// ((((expT * poly) + expT) + expH*poly) + expH)
-
-_CLC_DEF _CLC_OVERLOAD float __clc_pown(float x, int ny) {
- float y = (float)ny;
-
- int ix = __clc_as_int(x);
- int ax = ix & EXSIGNBIT_SP32;
- int xpos = ix == ax;
-
- int iy = __clc_as_int(y);
- int ay = iy & EXSIGNBIT_SP32;
- int ypos = iy == ay;
-
- // Extra precise log calculation
- // First handle case that x is close to 1
- float r = 1.0f - __clc_as_float(ax);
- int near1 = __clc_fabs(r) < 0x1.0p-4f;
- float r2 = r * r;
-
- // Coefficients are just 1/3, 1/4, 1/5 and 1/6
- float poly = __clc_mad(
- r,
- __clc_mad(r,
- __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
- 0x1.99999ap-3f),
- 0x1.000000p-2f),
- 0x1.555556p-2f);
-
- poly *= r2 * r;
-
- float lth_near1 = -r2 * 0.5f;
- float ltt_near1 = -poly;
- float lt_near1 = lth_near1 + ltt_near1;
- float lh_near1 = -r;
- float l_near1 = lh_near1 + lt_near1;
-
- // Computations for x not near 1
- int m = (int)(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
- float mf = (float)m;
- int ixs = __clc_as_int(__clc_as_float(ax | 0x3f800000) - 1.0f);
- float mfs = (float)((ixs >> EXPSHIFTBITS_SP32) - 253);
- int c = m == -127;
- int ixn = c ? ixs : ax;
- float mfn = c ? mfs : mf;
-
- int indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
-
- // F - Y
- float f = __clc_as_float(0x3f000000 | indx) -
- __clc_as_float(0x3f000000 | (ixn & MANTBITS_SP32));
-
- indx = indx >> 16;
- float2 tv = USE_TABLE(log_inv_tbl_ep, indx);
- float rh = f * tv.s0;
- float rt = f * tv.s1;
- r = rh + rt;
-
- poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
- (r * r);
- poly += (rh - r) + rt;
-
- const float LOG2_HEAD = 0x1.62e000p-1f; // 0.693115234
- const float LOG2_TAIL = 0x1.0bfbe8p-15f; // 0.0000319461833
- tv.s0 = USE_TABLE(loge_tbl_lo, indx);
- tv.s1 = USE_TABLE(loge_tbl_hi, indx);
- float lth = -r;
- float ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + tv.s1;
- float lt = lth + ltt;
- float lh = __clc_mad(mfn, LOG2_HEAD, tv.s0);
- float l = lh + lt;
-
- // Select near 1 or not
- lth = near1 ? lth_near1 : lth;
- ltt = near1 ? ltt_near1 : ltt;
- lt = near1 ? lt_near1 : lt;
- lh = near1 ? lh_near1 : lh;
- l = near1 ? l_near1 : l;
-
- float gh = __clc_as_float(__clc_as_int(l) & 0xfffff000);
- float gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
-
- float yh = __clc_as_float(iy & 0xfffff000);
-
- float yt = (float)(ny - (int)yh);
-
- float ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
- float ylogx = __clc_mad(yh, gh, ylogx_s);
- float ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
-
- // Extra precise exp of ylogx
- // 64/log2 : 92.332482616893657
- const float R_64_BY_LOG2 = 0x1.715476p+6f;
- int n = __clc_convert_int(ylogx * R_64_BY_LOG2);
- float nf = (float)n;
-
- int j = n & 0x3f;
- m = n >> 6;
- int m2 = m << EXPSHIFTBITS_SP32;
-
- // log2/64 lead: 0.0108032227
- const float R_LOG2_BY_64_LD = 0x1.620000p-7f;
- // log2/64 tail: 0.0000272020388
- const float R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
- r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
- ylogx_t;
-
- // Truncated Taylor series for e^r
- poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
- 0x1.000000p-1f),
- r * r, r);
-
- tv = USE_TABLE(exp_tbl_ep, j);
-
- float expylogx =
- __clc_mad(tv.s0, poly, __clc_mad(tv.s1, poly, tv.s1)) + tv.s0;
- float sexpylogx = expylogx * __clc_as_float(0x1 << (m + 149));
- float texpylogx = __clc_as_float(__clc_as_int(expylogx) + m2);
- expylogx = m < -125 ? sexpylogx : texpylogx;
-
- // Result is +-Inf if (ylogx + ylogx_t) > 128*log2
- expylogx = ((ylogx > 0x1.62e430p+6f) |
- (ylogx == 0x1.62e430p+6f & ylogx_t > -0x1.05c610p-22f))
- ? __clc_as_float(PINFBITPATT_SP32)
- : expylogx;
-
- // Result is 0 if ylogx < -149*log2
- expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
-
- // Classify y:
- // inty = 0 means not an integer.
- // inty = 1 means odd integer.
- // inty = 2 means even integer.
-
- int inty = 2 - (ny & 1);
-
- float signval = __clc_as_float((__clc_as_uint(expylogx) ^ SIGNBIT_SP32));
- expylogx = ((inty == 1) & !xpos) ? signval : expylogx;
- int ret = __clc_as_int(expylogx);
-
- // Corner case handling
- int xinf = xpos ? PINFBITPATT_SP32 : NINFBITPATT_SP32;
- ret = ((ax == 0) & !ypos & (inty == 1)) ? xinf : ret;
- ret = ((ax == 0) & !ypos & (inty == 2)) ? PINFBITPATT_SP32 : ret;
- ret = ((ax == 0) & ypos & (inty == 2)) ? 0 : ret;
- int xzero = !xpos ? 0x80000000 : 0L;
- ret = ((ax == 0) & ypos & (inty == 1)) ? xzero : ret;
- ret = ((ix == NINFBITPATT_SP32) & !ypos & (inty == 1)) ? 0x80000000 : ret;
- ret = ((ix == NINFBITPATT_SP32) & !ypos & (inty != 1)) ? 0 : ret;
- ret =
- ((ix == NINFBITPATT_SP32) & ypos & (inty == 1)) ? NINFBITPATT_SP32 : ret;
- ret =
- ((ix == NINFBITPATT_SP32) & ypos & (inty != 1)) ? PINFBITPATT_SP32 : ret;
- ret = ((ix == PINFBITPATT_SP32) & !ypos) ? 0 : ret;
- ret = ((ix == PINFBITPATT_SP32) & ypos) ? PINFBITPATT_SP32 : ret;
- ret = ax > PINFBITPATT_SP32 ? ix : ret;
- ret = ny == 0 ? 0x3f800000 : ret;
-
- return __clc_as_float(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, float, __clc_pown, float, int)
-
-#ifdef cl_khr_fp64
-_CLC_DEF _CLC_OVERLOAD double __clc_pown(double x, int ny) {
- const double real_log2_tail = 5.76999904754328540596e-08;
- const double real_log2_lead = 6.93147122859954833984e-01;
-
- double y = (double)ny;
-
- long ux = __clc_as_long(x);
- long ax = ux & (~SIGNBIT_DP64);
- int xpos = ax == ux;
-
- long uy = __clc_as_long(y);
- long ay = uy & (~SIGNBIT_DP64);
- int ypos = ay == uy;
-
- // Extended precision log
- double v, vt;
- {
- int exp = (int)(ax >> 52) - 1023;
- int mask_exp_1023 = exp == -1023;
- double xexp = (double)exp;
- long mantissa = ax & 0x000FFFFFFFFFFFFFL;
-
- long temp_ux =
- __clc_as_long(__clc_as_double(0x3ff0000000000000L | mantissa) - 1.0);
- exp = ((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
- double xexp1 = (double)exp;
- long mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
-
- xexp = mask_exp_1023 ? xexp1 : xexp;
- mantissa = mask_exp_1023 ? mantissa1 : mantissa;
-
- long rax = (mantissa & 0x000ff00000000000) +
- ((mantissa & 0x0000080000000000) << 1);
- int index = rax >> 44;
-
- double F = __clc_as_double(rax | 0x3FE0000000000000L);
- double Y = __clc_as_double(mantissa | 0x3FE0000000000000L);
- double f = F - Y;
- double2 tv = USE_TABLE(log_f_inv_tbl, index);
- double log_h = tv.s0;
- double log_t = tv.s1;
- double f_inv = (log_h + log_t) * f;
- double r1 = __clc_as_double(__clc_as_long(f_inv) & 0xfffffffff8000000L);
- double r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
- double r = r1 + r2;
-
- double poly = __clc_fma(
- r,
- __clc_fma(r,
- __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
- 1.0 / 4.0),
- 1.0 / 3.0);
- poly = poly * r * r * r;
-
- double hr1r1 = 0.5 * r1 * r1;
- double poly0h = r1 + hr1r1;
- double poly0t = r1 - poly0h + hr1r1;
- poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
-
- tv = USE_TABLE(powlog_tbl, index);
- log_h = tv.s0;
- log_t = tv.s1;
-
- double resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
- double resT = resT_t - poly0h;
- double resH = __clc_fma(xexp, real_log2_lead, log_h);
- double resT_h = poly0h;
-
- double H = resT + resH;
- double H_h = __clc_as_double(__clc_as_long(H) & 0xfffffffff8000000L);
- double T = (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
- H = H_h;
-
- double y_head = __clc_as_double(uy & 0xfffffffff8000000L);
- double y_tail = y - y_head;
-
- int mask_2_24 = ay > 0x4170000000000000; // 2^24
- int nyh = __clc_convert_int(y_head);
- int nyt = ny - nyh;
- double y_tail1 = (double)nyt;
- y_tail = mask_2_24 ? y_tail1 : y_tail;
-
- double temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
- v = __clc_fma(y_head, H, temp);
- vt = __clc_fma(y_head, H, -v) + temp;
- }
-
- // Now calculate exp of (v,vt)
-
- double expv;
- {
- const double max_exp_arg = 709.782712893384;
- const double min_exp_arg = -745.1332191019411;
- const double sixtyfour_by_lnof2 = 92.33248261689366;
- const double lnof2_by_64_head = 0.010830424260348081;
- const double lnof2_by_64_tail = -4.359010638708991e-10;
-
- double temp = v * sixtyfour_by_lnof2;
- int n = (int)temp;
- double dn = (double)n;
- int j = n & 0x0000003f;
- int m = n >> 6;
-
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- double f1 = tv.s0;
- double f2 = tv.s1;
- double f = f1 + f2;
-
- double r1 = __clc_fma(dn, -lnof2_by_64_head, v);
- double r2 = dn * lnof2_by_64_tail;
- double r = (r1 + r2) + vt;
-
- double q =
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r, 1.38889490863777199667e-03,
- 8.33336798434219616221e-03),
- 4.16666666662260795726e-02),
- 1.66666666665260878863e-01),
- 5.00000000000000008883e-01);
- q = __clc_fma(r * r, q, r);
-
- expv = __clc_fma(f, q, f2) + f1;
- expv = ldexp(expv, m);
-
- expv = v > max_exp_arg ? __clc_as_double(0x7FF0000000000000L) : expv;
- expv = v < min_exp_arg ? 0.0 : expv;
- }
-
- // See whether y is an integer.
- // inty = 0 means not an integer.
- // inty = 1 means odd integer.
- // inty = 2 means even integer.
-
- int inty = 2 - (ny & 1);
-
- expv *= ((inty == 1) & !xpos) ? -1.0 : 1.0;
-
- long ret = __clc_as_long(expv);
-
- // Now all the edge cases
- long xinf = xpos ? PINFBITPATT_DP64 : NINFBITPATT_DP64;
- ret = ((ax == 0L) & !ypos & (inty == 1)) ? xinf : ret;
- ret = ((ax == 0L) & !ypos & (inty == 2)) ? PINFBITPATT_DP64 : ret;
- ret = ((ax == 0L) & ypos & (inty == 2)) ? 0L : ret;
- long xzero = !xpos ? 0x8000000000000000L : 0L;
- ret = ((ax == 0L) & ypos & (inty == 1)) ? xzero : ret;
- ret = ((ux == NINFBITPATT_DP64) & !ypos & (inty == 1)) ? 0x8000000000000000L
- : ret;
- ret = ((ux == NINFBITPATT_DP64) & !ypos & (inty != 1)) ? 0L : ret;
- ret =
- ((ux == NINFBITPATT_DP64) & ypos & (inty == 1)) ? NINFBITPATT_DP64 : ret;
- ret =
- ((ux == NINFBITPATT_DP64) & ypos & (inty != 1)) ? PINFBITPATT_DP64 : ret;
- ret = ((ux == PINFBITPATT_DP64) & !ypos) ? 0L : ret;
- ret = ((ux == PINFBITPATT_DP64) & ypos) ? PINFBITPATT_DP64 : ret;
- ret = ax > PINFBITPATT_DP64 ? ux : ret;
- ret = ny == 0 ? 0x3ff0000000000000L : ret;
-
- return __clc_as_double(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, double, __clc_pown, double, int)
-#endif
-
-#ifdef cl_khr_fp16
-
-#pragma OPENCL EXTENSION cl_khr_fp16 : enable
-
-_CLC_OVERLOAD _CLC_DEF half __clc_pown(half x, int y) {
- return (half)__clc_pown((float)x, y);
-}
-
-_CLC_BINARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, half, __clc_pown, half, int);
-
-#endif
diff --git a/libclc/generic/lib/math/clc_powr.cl b/libclc/generic/lib/math/clc_powr.cl
deleted file mode 100644
index 428101cf78bc5..0000000000000
--- a/libclc/generic/lib/math/clc_powr.cl
+++ /dev/null
@@ -1,390 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
-// See https://llvm.org/LICENSE.txt for license information.
-// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
-//
-//===----------------------------------------------------------------------===//
-
-#include <clc/clc.h>
-#include <clc/clc_convert.h>
-#include <clc/clcmacro.h>
-#include <clc/math/clc_fabs.h>
-#include <clc/math/clc_fma.h>
-#include <clc/math/clc_mad.h>
-#include <clc/math/clc_subnormal_config.h>
-#include <clc/math/math.h>
-#include <clc/math/tables.h>
-
-// compute pow using log and exp
-// x^y = exp(y * log(x))
-//
-// we take care not to lose precision in the intermediate steps
-//
-// When computing log, calculate it in splits,
-//
-// r = f * (p_invead + p_inv_tail)
-// r = rh + rt
-//
-// calculate log polynomial using r, in end addition, do
-// poly = poly + ((rh-r) + rt)
-//
-// lth = -r
-// ltt = ((xexp * log2_t) - poly) + logT
-// lt = lth + ltt
-//
-// lh = (xexp * log2_h) + logH
-// l = lh + lt
-//
-// Calculate final log answer as gh and gt,
-// gh = l & higher-half bits
-// gt = (((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh))
-//
-// yh = y & higher-half bits
-// yt = y - yh
-//
-// Before entering computation of exp,
-// vs = ((yt*gt + yt*gh) + yh*gt)
-// v = vs + yh*gh
-// vt = ((yh*gh - v) + vs)
-//
-// In calculation of exp, add vt to r that is used for poly
-// At the end of exp, do
-// ((((expT * poly) + expT) + expH*poly) + expH)
-
-_CLC_DEF _CLC_OVERLOAD float __clc_powr(float x, float y) {
- int ix = __clc_as_int(x);
- int ax = ix & EXSIGNBIT_SP32;
- int xpos = ix == ax;
-
- int iy = __clc_as_int(y);
- int ay = iy & EXSIGNBIT_SP32;
- int ypos = iy == ay;
-
- // Extra precise log calculation
- // First handle case that x is close to 1
- float r = 1.0f - __clc_as_float(ax);
- int near1 = __clc_fabs(r) < 0x1.0p-4f;
- float r2 = r * r;
-
- // Coefficients are just 1/3, 1/4, 1/5 and 1/6
- float poly = __clc_mad(
- r,
- __clc_mad(r,
- __clc_mad(r, __clc_mad(r, 0x1.24924ap-3f, 0x1.555556p-3f),
- 0x1.99999ap-3f),
- 0x1.000000p-2f),
- 0x1.555556p-2f);
-
- poly *= r2 * r;
-
- float lth_near1 = -r2 * 0.5f;
- float ltt_near1 = -poly;
- float lt_near1 = lth_near1 + ltt_near1;
- float lh_near1 = -r;
- float l_near1 = lh_near1 + lt_near1;
-
- // Computations for x not near 1
- int m = (int)(ax >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32;
- float mf = (float)m;
- int ixs = __clc_as_int(__clc_as_float(ax | 0x3f800000) - 1.0f);
- float mfs = (float)((ixs >> EXPSHIFTBITS_SP32) - 253);
- int c = m == -127;
- int ixn = c ? ixs : ax;
- float mfn = c ? mfs : mf;
-
- int indx = (ixn & 0x007f0000) + ((ixn & 0x00008000) << 1);
-
- // F - Y
- float f = __clc_as_float(0x3f000000 | indx) -
- __clc_as_float(0x3f000000 | (ixn & MANTBITS_SP32));
-
- indx = indx >> 16;
- float2 tv = USE_TABLE(log_inv_tbl_ep, indx);
- float rh = f * tv.s0;
- float rt = f * tv.s1;
- r = rh + rt;
-
- poly = __clc_mad(r, __clc_mad(r, 0x1.0p-2f, 0x1.555556p-2f), 0x1.0p-1f) *
- (r * r);
- poly += (rh - r) + rt;
-
- const float LOG2_HEAD = 0x1.62e000p-1f; // 0.693115234
- const float LOG2_TAIL = 0x1.0bfbe8p-15f; // 0.0000319461833
- tv.s0 = USE_TABLE(loge_tbl_lo, indx);
- tv.s1 = USE_TABLE(loge_tbl_hi, indx);
- float lth = -r;
- float ltt = __clc_mad(mfn, LOG2_TAIL, -poly) + tv.s1;
- float lt = lth + ltt;
- float lh = __clc_mad(mfn, LOG2_HEAD, tv.s0);
- float l = lh + lt;
-
- // Select near 1 or not
- lth = near1 ? lth_near1 : lth;
- ltt = near1 ? ltt_near1 : ltt;
- lt = near1 ? lt_near1 : lt;
- lh = near1 ? lh_near1 : lh;
- l = near1 ? l_near1 : l;
-
- float gh = __clc_as_float(__clc_as_int(l) & 0xfffff000);
- float gt = ((ltt - (lt - lth)) + ((lh - l) + lt)) + (l - gh);
-
- float yh = __clc_as_float(iy & 0xfffff000);
-
- float yt = y - yh;
-
- float ylogx_s = __clc_mad(gt, yh, __clc_mad(gh, yt, yt * gt));
- float ylogx = __clc_mad(yh, gh, ylogx_s);
- float ylogx_t = __clc_mad(yh, gh, -ylogx) + ylogx_s;
-
- // Extra precise exp of ylogx
- // 64/log2 : 92.332482616893657
- const float R_64_BY_LOG2 = 0x1.715476p+6f;
- int n = __clc_convert_int(ylogx * R_64_BY_LOG2);
- float nf = (float)n;
-
- int j = n & 0x3f;
- m = n >> 6;
- int m2 = m << EXPSHIFTBITS_SP32;
- // log2/64 lead: 0.0108032227
- const float R_LOG2_BY_64_LD = 0x1.620000p-7f;
- // log2/64 tail: 0.0000272020388
- const float R_LOG2_BY_64_TL = 0x1.c85fdep-16f;
- r = __clc_mad(nf, -R_LOG2_BY_64_TL, __clc_mad(nf, -R_LOG2_BY_64_LD, ylogx)) +
- ylogx_t;
-
- // Truncated Taylor series for e^r
- poly = __clc_mad(__clc_mad(__clc_mad(r, 0x1.555556p-5f, 0x1.555556p-3f), r,
- 0x1.000000p-1f),
- r * r, r);
-
- tv = USE_TABLE(exp_tbl_ep, j);
-
- float expylogx =
- __clc_mad(tv.s0, poly, __clc_mad(tv.s1, poly, tv.s1)) + tv.s0;
- float sexpylogx = expylogx * __clc_as_float(0x1 << (m + 149));
- float texpylogx = __clc_as_float(__clc_as_int(expylogx) + m2);
- expylogx = m < -125 ? sexpylogx : texpylogx;
-
- // Result is +-Inf if (ylogx + ylogx_t) > 128*log2
- expylogx = ((ylogx > 0x1.62e430p+6f) |
- (ylogx == 0x1.62e430p+6f & ylogx_t > -0x1.05c610p-22f))
- ? __clc_as_float(PINFBITPATT_SP32)
- : expylogx;
-
- // Result is 0 if ylogx < -149*log2
- expylogx = ylogx < -0x1.9d1da0p+6f ? 0.0f : expylogx;
-
- // Classify y:
- // inty = 0 means not an integer.
- // inty = 1 means odd integer.
- // inty = 2 means even integer.
-
- int yexp = (int)(ay >> EXPSHIFTBITS_SP32) - EXPBIAS_SP32 + 1;
- int mask = (1 << (24 - yexp)) - 1;
- int yodd = ((iy >> (24 - yexp)) & 0x1) != 0;
- int inty = yodd ? 1 : 2;
- inty = (iy & mask) != 0 ? 0 : inty;
- inty = yexp < 1 ? 0 : inty;
- inty = yexp > 24 ? 2 : inty;
-
- float signval = __clc_as_float((__clc_as_uint(expylogx) ^ SIGNBIT_SP32));
- expylogx = ((inty == 1) & !xpos) ? signval : expylogx;
- int ret = __clc_as_int(expylogx);
-
- // Corner case handling
- ret = ax < 0x3f800000 & iy == NINFBITPATT_SP32 ? PINFBITPATT_SP32 : ret;
- ret = ax < 0x3f800000 & iy == PINFBITPATT_SP32 ? 0 : ret;
- ret = ax == 0x3f800000 & ay < PINFBITPATT_SP32 ? 0x3f800000 : ret;
- ret = ax == 0x3f800000 & ay == PINFBITPATT_SP32 ? QNANBITPATT_SP32 : ret;
- ret = ax > 0x3f800000 & iy == NINFBITPATT_SP32 ? 0 : ret;
- ret = ax > 0x3f800000 & iy == PINFBITPATT_SP32 ? PINFBITPATT_SP32 : ret;
- ret = ((ix < PINFBITPATT_SP32) & (ay == 0)) ? 0x3f800000 : ret;
- ret = ((ax == PINFBITPATT_SP32) & !ypos) ? 0 : ret;
- ret = ((ax == PINFBITPATT_SP32) & ypos) ? PINFBITPATT_SP32 : ret;
- ret = ((ax == PINFBITPATT_SP32) & (iy == PINFBITPATT_SP32)) ? PINFBITPATT_SP32
- : ret;
- ret = ((ax == PINFBITPATT_SP32) & (ay == 0)) ? QNANBITPATT_SP32 : ret;
- ret = ((ax == 0) & !ypos) ? PINFBITPATT_SP32 : ret;
- ret = ((ax == 0) & ypos) ? 0 : ret;
- ret = ((ax == 0) & (ay == 0)) ? QNANBITPATT_SP32 : ret;
- ret = ((ax != 0) & !xpos) ? QNANBITPATT_SP32 : ret;
- ret = ax > PINFBITPATT_SP32 ? ix : ret;
- ret = ay > PINFBITPATT_SP32 ? iy : ret;
-
- return __clc_as_float(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, float, __clc_powr, float, float)
-
-#ifdef cl_khr_fp64
-_CLC_DEF _CLC_OVERLOAD double __clc_powr(double x, double y) {
- const double real_log2_tail = 5.76999904754328540596e-08;
- const double real_log2_lead = 6.93147122859954833984e-01;
-
- long ux = __clc_as_long(x);
- long ax = ux & (~SIGNBIT_DP64);
- int xpos = ax == ux;
-
- long uy = __clc_as_long(y);
- long ay = uy & (~SIGNBIT_DP64);
- int ypos = ay == uy;
-
- // Extended precision log
- double v, vt;
- {
- int exp = (int)(ax >> 52) - 1023;
- int mask_exp_1023 = exp == -1023;
- double xexp = (double)exp;
- long mantissa = ax & 0x000FFFFFFFFFFFFFL;
-
- long temp_ux =
- __clc_as_long(__clc_as_double(0x3ff0000000000000L | mantissa) - 1.0);
- exp = ((temp_ux & 0x7FF0000000000000L) >> 52) - 2045;
- double xexp1 = (double)exp;
- long mantissa1 = temp_ux & 0x000FFFFFFFFFFFFFL;
-
- xexp = mask_exp_1023 ? xexp1 : xexp;
- mantissa = mask_exp_1023 ? mantissa1 : mantissa;
-
- long rax = (mantissa & 0x000ff00000000000) +
- ((mantissa & 0x0000080000000000) << 1);
- int index = rax >> 44;
-
- double F = __clc_as_double(rax | 0x3FE0000000000000L);
- double Y = __clc_as_double(mantissa | 0x3FE0000000000000L);
- double f = F - Y;
- double2 tv = USE_TABLE(log_f_inv_tbl, index);
- double log_h = tv.s0;
- double log_t = tv.s1;
- double f_inv = (log_h + log_t) * f;
- double r1 = __clc_as_double(__clc_as_long(f_inv) & 0xfffffffff8000000L);
- double r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
- double r = r1 + r2;
-
- double poly = __clc_fma(
- r,
- __clc_fma(r,
- __clc_fma(r, __clc_fma(r, 1.0 / 7.0, 1.0 / 6.0), 1.0 / 5.0),
- 1.0 / 4.0),
- 1.0 / 3.0);
- poly = poly * r * r * r;
-
- double hr1r1 = 0.5 * r1 * r1;
- double poly0h = r1 + hr1r1;
- double poly0t = r1 - poly0h + hr1r1;
- poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
-
- tv = USE_TABLE(powlog_tbl, index);
- log_h = tv.s0;
- log_t = tv.s1;
-
- double resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
- double resT = resT_t - poly0h;
- double resH = __clc_fma(xexp, real_log2_lead, log_h);
- double resT_h = poly0h;
-
- double H = resT + resH;
- double H_h = __clc_as_double(__clc_as_long(H) & 0xfffffffff8000000L);
- double T = (resH - H + resT) + (resT_t - (resT + resT_h)) + (H - H_h);
- H = H_h;
-
- double y_head = __clc_as_double(uy & 0xfffffffff8000000L);
- double y_tail = y - y_head;
-
- double temp = __clc_fma(y_tail, H, __clc_fma(y_head, T, y_tail * T));
- v = __clc_fma(y_head, H, temp);
- vt = __clc_fma(y_head, H, -v) + temp;
- }
-
- // Now calculate exp of (v,vt)
-
- double expv;
- {
- const double max_exp_arg = 709.782712893384;
- const double min_exp_arg = -745.1332191019411;
- const double sixtyfour_by_lnof2 = 92.33248261689366;
- const double lnof2_by_64_head = 0.010830424260348081;
- const double lnof2_by_64_tail = -4.359010638708991e-10;
-
- double temp = v * sixtyfour_by_lnof2;
- int n = (int)temp;
- double dn = (double)n;
- int j = n & 0x0000003f;
- int m = n >> 6;
-
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- double f1 = tv.s0;
- double f2 = tv.s1;
- double f = f1 + f2;
-
- double r1 = __clc_fma(dn, -lnof2_by_64_head, v);
- double r2 = dn * lnof2_by_64_tail;
- double r = (r1 + r2) + vt;
-
- double q =
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r,
- __clc_fma(r, 1.38889490863777199667e-03,
- 8.33336798434219616221e-03),
- 4.16666666662260795726e-02),
- 1.66666666665260878863e-01),
- 5.00000000000000008883e-01);
- q = __clc_fma(r * r, q, r);
-
- expv = __clc_fma(f, q, f2) + f1;
- expv = ldexp(expv, m);
-
- expv = v > max_exp_arg ? __clc_as_double(0x7FF0000000000000L) : expv;
- expv = v < min_exp_arg ? 0.0 : expv;
- }
-
- // See whether y is an integer.
- // inty = 0 means not an integer.
- // inty = 1 means odd integer.
- // inty = 2 means even integer.
-
- int inty;
- {
- int yexp = (int)(ay >> EXPSHIFTBITS_DP64) - EXPBIAS_DP64 + 1;
- inty = yexp < 1 ? 0 : 2;
- inty = yexp > 53 ? 2 : inty;
- long mask = (1L << (53 - yexp)) - 1L;
- int inty1 = (((ay & ~mask) >> (53 - yexp)) & 1L) == 1L ? 1 : 2;
- inty1 = (ay & mask) != 0 ? 0 : inty1;
- inty = !(yexp < 1) & !(yexp > 53) ? inty1 : inty;
- }
-
- expv *= ((inty == 1) & !xpos) ? -1.0 : 1.0;
-
- long ret = __clc_as_long(expv);
-
- // Now all the edge cases
- ret = ax < 0x3ff0000000000000L & uy == NINFBITPATT_DP64 ? PINFBITPATT_DP64
- : ret;
- ret = ax < 0x3ff0000000000000L & uy == PINFBITPATT_DP64 ? 0L : ret;
- ret = ax == 0x3ff0000000000000L & ay < PINFBITPATT_DP64 ? 0x3ff0000000000000L
- : ret;
- ret = ax == 0x3ff0000000000000L & ay == PINFBITPATT_DP64 ? QNANBITPATT_DP64
- : ret;
- ret = ax > 0x3ff0000000000000L & uy == NINFBITPATT_DP64 ? 0L : ret;
- ret = ax > 0x3ff0000000000000L & uy == PINFBITPATT_DP64 ? PINFBITPATT_DP64
- : ret;
- ret = ux < PINFBITPATT_DP64 & ay == 0L ? 0x3ff0000000000000L : ret;
- ret = ((ax == PINFBITPATT_DP64) & !ypos) ? 0L : ret;
- ret = ((ax == PINFBITPATT_DP64) & ypos) ? PINFBITPATT_DP64 : ret;
- ret = ((ax == PINFBITPATT_DP64) & (uy == PINFBITPATT_DP64)) ? PINFBITPATT_DP64
- : ret;
- ret = ((ax == PINFBITPATT_DP64) & (ay == 0L)) ? QNANBITPATT_DP64 : ret;
- ret = ((ax == 0L) & !ypos) ? PINFBITPATT_DP64 : ret;
- ret = ((ax == 0L) & ypos) ? 0L : ret;
- ret = ((ax == 0L) & (ay == 0L)) ? QNANBITPATT_DP64 : ret;
- ret = ((ax != 0L) & !xpos) ? QNANBITPATT_DP64 : ret;
- ret = ax > PINFBITPATT_DP64 ? ux : ret;
- ret = ay > PINFBITPATT_DP64 ? uy : ret;
-
- return __clc_as_double(ret);
-}
-_CLC_BINARY_VECTORIZE(_CLC_DEF _CLC_OVERLOAD, double, __clc_powr, double,
- double)
-#endif
diff --git a/libclc/generic/lib/math/clc_rootn.cl b/libclc/generic/lib/math/clc_rootn.cl
index 99fe539dad26d..f642792e1275f 100644
--- a/libclc/generic/lib/math/clc_rootn.cl
+++ b/libclc/generic/lib/math/clc_rootn.cl
@@ -102,7 +102,9 @@ _CLC_DEF _CLC_OVERLOAD float __clc_rootn(float x, int ny) {
__clc_as_float(0x3f000000 | (ixn & MANTBITS_SP32));
indx = indx >> 16;
- float2 tv = USE_TABLE(log_inv_tbl_ep, indx);
+ float2 tv;
+ tv.s0 = USE_TABLE(log_inv_tbl_ep_head, indx);
+ tv.s1 = USE_TABLE(log_inv_tbl_ep_tail, indx);
float rh = f * tv.s0;
float rt = f * tv.s1;
r = rh + rt;
@@ -163,7 +165,8 @@ _CLC_DEF _CLC_OVERLOAD float __clc_rootn(float x, int ny) {
0x1.000000p-1f),
r * r, r);
- tv = USE_TABLE(exp_tbl_ep, j);
+ tv.s0 = USE_TABLE(exp_tbl_ep_head, j);
+ tv.s1 = USE_TABLE(exp_tbl_ep_tail, j);
float expylogx =
__clc_mad(tv.s0, poly, __clc_mad(tv.s1, poly, tv.s1)) + tv.s0;
@@ -254,9 +257,8 @@ _CLC_DEF _CLC_OVERLOAD double __clc_rootn(double x, int ny) {
double F = __clc_as_double(rax | 0x3FE0000000000000L);
double Y = __clc_as_double(mantissa | 0x3FE0000000000000L);
double f = F - Y;
- double2 tv = USE_TABLE(log_f_inv_tbl, index);
- double log_h = tv.s0;
- double log_t = tv.s1;
+ double log_h = USE_TABLE(log_f_inv_tbl_head, index);
+ double log_t = USE_TABLE(log_f_inv_tbl_tail, index);
double f_inv = (log_h + log_t) * f;
double r1 = __clc_as_double(__clc_as_long(f_inv) & 0xfffffffff8000000L);
double r2 = __clc_fma(-F, r1, f) * (log_h + log_t);
@@ -275,9 +277,8 @@ _CLC_DEF _CLC_OVERLOAD double __clc_rootn(double x, int ny) {
double poly0t = r1 - poly0h + hr1r1;
poly = __clc_fma(r1, r2, __clc_fma(0.5 * r2, r2, poly)) + r2 + poly0t;
- tv = USE_TABLE(powlog_tbl, index);
- log_h = tv.s0;
- log_t = tv.s1;
+ log_h = USE_TABLE(powlog_tbl_head, index);
+ log_t = USE_TABLE(powlog_tbl_tail, index);
double resT_t = __clc_fma(xexp, real_log2_tail, +log_t) - poly;
double resT = resT_t - poly0h;
@@ -317,9 +318,8 @@ _CLC_DEF _CLC_OVERLOAD double __clc_rootn(double x, int ny) {
int j = n & 0x0000003f;
int m = n >> 6;
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- double f1 = tv.s0;
- double f2 = tv.s1;
+ double f1 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ double f2 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
double f = f1 + f2;
double r1 = __clc_fma(dn, -lnof2_by_64_head, v);
diff --git a/libclc/generic/lib/math/exp_helper.cl b/libclc/generic/lib/math/exp_helper.cl
index a8ffec2034f8e..b413228719bfb 100644
--- a/libclc/generic/lib/math/exp_helper.cl
+++ b/libclc/generic/lib/math/exp_helper.cl
@@ -30,8 +30,9 @@ _CLC_DEF double __clc_exp_helper(double x, double x_min, double x_max, double r,
0x1.0000000000000p-1),
1.0);
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- z2 = fma(tv.s0 + tv.s1, z2, tv.s1) + tv.s0;
+ double tv0 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ double tv1 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+ z2 = fma(tv0 + tv1, z2, tv1) + tv0;
int small_value = (m < -1022) || ((m == -1022) && (z2 < 1.0));
diff --git a/libclc/generic/lib/math/expm1.cl b/libclc/generic/lib/math/expm1.cl
index cce6a42ee0124..e66020e20fab4 100644
--- a/libclc/generic/lib/math/expm1.cl
+++ b/libclc/generic/lib/math/expm1.cl
@@ -34,10 +34,11 @@ _CLC_OVERLOAD _CLC_DEF float expm1(float x) {
float z2 = mad(r*r, mad(r, mad(r, 0x1.555556p-5f, 0x1.555556p-3f), 0.5f), r);
float m2 = as_float((m + EXPBIAS_SP32) << EXPSHIFTBITS_SP32);
- float2 tv = USE_TABLE(exp_tbl_ep, j);
+ float exp_head = USE_TABLE(exp_tbl_ep_head, j);
+ float exp_tail = USE_TABLE(exp_tbl_ep_tail, j);
- float two_to_jby64_h = tv.s0 * m2;
- float two_to_jby64_t = tv.s1 * m2;
+ float two_to_jby64_h = exp_head * m2;
+ float two_to_jby64_t = exp_tail * m2;
float two_to_jby64 = two_to_jby64_h + two_to_jby64_t;
z2 = mad(z2, two_to_jby64, two_to_jby64_t) + (two_to_jby64_h - 1.0f);
@@ -99,9 +100,8 @@ _CLC_OVERLOAD _CLC_DEF double expm1(double x) {
int j = n & 0x3f;
int m = n >> 6;
- double2 tv = USE_TABLE(two_to_jby64_ep_tbl, j);
- double f1 = tv.s0;
- double f2 = tv.s1;
+ double f1 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+ double f2 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
double f = f1 + f2;
double dn = -n;
diff --git a/libclc/generic/lib/math/pow.cl b/libclc/generic/lib/math/pow.cl
index 68cb329e9775f..a997d32e16f5e 100644
--- a/libclc/generic/lib/math/pow.cl
+++ b/libclc/generic/lib/math/pow.cl
@@ -7,9 +7,8 @@
//===----------------------------------------------------------------------===//
#include <clc/clc.h>
+#include <clc/math/clc_pow.h>
-#include <math/clc_pow.h>
-
-#define __CLC_FUNC pow
-#define __CLC_BODY <clc_sw_binary.inc>
+#define FUNCTION pow
+#define __CLC_BODY <clc/shared/binary_def.inc>
#include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/math/pown.cl b/libclc/generic/lib/math/pown.cl
index 6b2d3b300666f..8c077f7aae148 100644
--- a/libclc/generic/lib/math/pown.cl
+++ b/libclc/generic/lib/math/pown.cl
@@ -7,8 +7,8 @@
//===----------------------------------------------------------------------===//
#include <clc/clc.h>
+#include <clc/math/clc_pown.h>
-#include <math/clc_pown.h>
-
-#define __CLC_BODY <pown.inc>
+#define FUNCTION pown
+#define __CLC_BODY <clc/math/binary_def_with_int_second_arg.inc>
#include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/math/powr.cl b/libclc/generic/lib/math/powr.cl
index 909ac18c9d76c..8b3df54a2434a 100644
--- a/libclc/generic/lib/math/powr.cl
+++ b/libclc/generic/lib/math/powr.cl
@@ -7,9 +7,8 @@
//===----------------------------------------------------------------------===//
#include <clc/clc.h>
+#include <clc/math/clc_powr.h>
-#include <math/clc_powr.h>
-
-#define __CLC_FUNC powr
-#define __CLC_BODY <clc_sw_binary.inc>
+#define FUNCTION powr
+#define __CLC_BODY <clc/shared/binary_def.inc>
#include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/math/tables.cl b/libclc/generic/lib/math/tables.cl
index c03b4d1a3c18a..5534ae36d4e5b 100644
--- a/libclc/generic/lib/math/tables.cl
+++ b/libclc/generic/lib/math/tables.cl
@@ -10,138 +10,6 @@
#include <clc/math/tables.h>
-DECLARE_TABLE(float2, LOG_INV_TBL_EP, 129) = {
- (float2)(0x1.000000p+1f, 0x0.000000p+0f),
- (float2)(0x1.fc0000p+0f, 0x1.fc07f0p-14f),
- (float2)(0x1.f80000p+0f, 0x1.f81f82p-12f),
- (float2)(0x1.f40000p+0f, 0x1.196792p-10f),
- (float2)(0x1.f00000p+0f, 0x1.f07c20p-10f),
- (float2)(0x1.ec0000p+0f, 0x1.80f660p-9f),
- (float2)(0x1.e80000p+0f, 0x1.131ac0p-8f),
- (float2)(0x1.e40000p+0f, 0x1.73ac90p-8f),
- (float2)(0x1.e00000p+0f, 0x1.e1e1e2p-8f),
- (float2)(0x1.de0000p+0f, 0x1.75b8fep-10f),
- (float2)(0x1.da0000p+0f, 0x1.cc0ed8p-9f),
- (float2)(0x1.d60000p+0f, 0x1.7b654cp-8f),
- (float2)(0x1.d40000p+0f, 0x1.d41d42p-12f),
- (float2)(0x1.d00000p+0f, 0x1.96b1eep-9f),
- (float2)(0x1.cc0000p+0f, 0x1.856890p-8f),
- (float2)(0x1.ca0000p+0f, 0x1.2cc158p-10f),
- (float2)(0x1.c60000p+0f, 0x1.1c71c8p-8f),
- (float2)(0x1.c20000p+0f, 0x1.f8f01cp-8f),
- (float2)(0x1.c00000p+0f, 0x1.c0e070p-9f),
- (float2)(0x1.bc0000p+0f, 0x1.d2b89ap-8f),
- (float2)(0x1.ba0000p+0f, 0x1.9f2298p-9f),
- (float2)(0x1.b60000p+0f, 0x1.d6c3dep-8f),
- (float2)(0x1.b40000p+0f, 0x1.d0369ep-9f),
- (float2)(0x1.b20000p+0f, 0x1.b20364p-15f),
- (float2)(0x1.ae0000p+0f, 0x1.286bcap-8f),
- (float2)(0x1.ac0000p+0f, 0x1.5c06b2p-10f),
- (float2)(0x1.a80000p+0f, 0x1.8ef606p-8f),
- (float2)(0x1.a60000p+0f, 0x1.a034dap-9f),
- (float2)(0x1.a40000p+0f, 0x1.a41a42p-12f),
- (float2)(0x1.a00000p+0f, 0x1.6d3f98p-8f),
- (float2)(0x1.9e0000p+0f, 0x1.91d2a2p-9f),
- (float2)(0x1.9c0000p+0f, 0x1.68a772p-11f),
- (float2)(0x1.980000p+0f, 0x1.99999ap-8f),
- (float2)(0x1.960000p+0f, 0x1.0e4f80p-8f),
- (float2)(0x1.940000p+0f, 0x1.161f9ap-9f),
- (float2)(0x1.920000p+0f, 0x1.f693a2p-13f),
- (float2)(0x1.8e0000p+0f, 0x1.9c18fap-8f),
- (float2)(0x1.8c0000p+0f, 0x1.3018d4p-8f),
- (float2)(0x1.8a0000p+0f, 0x1.9721eep-9f),
- (float2)(0x1.880000p+0f, 0x1.b97c2ap-10f),
- (float2)(0x1.860000p+0f, 0x1.861862p-12f),
- (float2)(0x1.820000p+0f, 0x1.c977acp-8f),
- (float2)(0x1.800000p+0f, 0x1.818182p-8f),
- (float2)(0x1.7e0000p+0f, 0x1.405fd0p-8f),
- (float2)(0x1.7c0000p+0f, 0x1.05f418p-8f),
- (float2)(0x1.7a0000p+0f, 0x1.a4411cp-9f),
- (float2)(0x1.780000p+0f, 0x1.499030p-9f),
- (float2)(0x1.760000p+0f, 0x1.f7390ep-10f),
- (float2)(0x1.740000p+0f, 0x1.745d18p-10f),
- (float2)(0x1.720000p+0f, 0x1.0a1fd2p-10f),
- (float2)(0x1.700000p+0f, 0x1.702e06p-11f),
- (float2)(0x1.6e0000p+0f, 0x1.f76b44p-12f),
- (float2)(0x1.6c0000p+0f, 0x1.6c16c2p-12f),
- (float2)(0x1.6a0000p+0f, 0x1.3cd154p-12f),
- (float2)(0x1.680000p+0f, 0x1.681682p-12f),
- (float2)(0x1.660000p+0f, 0x1.ec6a52p-12f),
- (float2)(0x1.640000p+0f, 0x1.642c86p-11f),
- (float2)(0x1.620000p+0f, 0x1.fd3b80p-11f),
- (float2)(0x1.600000p+0f, 0x1.605816p-10f),
- (float2)(0x1.5e0000p+0f, 0x1.d6ee34p-10f),
- (float2)(0x1.5c0000p+0f, 0x1.310572p-9f),
- (float2)(0x1.5a0000p+0f, 0x1.80ad60p-9f),
- (float2)(0x1.580000p+0f, 0x1.da4610p-9f),
- (float2)(0x1.560000p+0f, 0x1.1ed3c6p-8f),
- (float2)(0x1.540000p+0f, 0x1.555556p-8f),
- (float2)(0x1.520000p+0f, 0x1.909490p-8f),
- (float2)(0x1.500000p+0f, 0x1.d07eaep-8f),
- (float2)(0x1.500000p+0f, 0x1.501502p-12f),
- (float2)(0x1.4e0000p+0f, 0x1.7829ccp-10f),
- (float2)(0x1.4c0000p+0f, 0x1.5710e4p-9f),
- (float2)(0x1.4a0000p+0f, 0x1.fad40ap-9f),
- (float2)(0x1.480000p+0f, 0x1.539e3cp-8f),
- (float2)(0x1.460000p+0f, 0x1.ae147ap-8f),
- (float2)(0x1.460000p+0f, 0x1.978fecp-13f),
- (float2)(0x1.440000p+0f, 0x1.be1958p-10f),
- (float2)(0x1.420000p+0f, 0x1.acc4bap-9f),
- (float2)(0x1.400000p+0f, 0x1.414142p-8f),
- (float2)(0x1.3e0000p+0f, 0x1.b013fcp-8f),
- (float2)(0x1.3e0000p+0f, 0x1.165e72p-11f),
- (float2)(0x1.3c0000p+0f, 0x1.32b490p-9f),
- (float2)(0x1.3a0000p+0f, 0x1.13b13cp-8f),
- (float2)(0x1.380000p+0f, 0x1.91c2c2p-8f),
- (float2)(0x1.380000p+0f, 0x1.381382p-12f),
- (float2)(0x1.360000p+0f, 0x1.31be7cp-9f),
- (float2)(0x1.340000p+0f, 0x1.21cfb2p-8f),
- (float2)(0x1.320000p+0f, 0x1.ae45b6p-8f),
- (float2)(0x1.320000p+0f, 0x1.f1a516p-11f),
- (float2)(0x1.300000p+0f, 0x1.a32026p-9f),
- (float2)(0x1.2e0000p+0f, 0x1.684bdap-8f),
- (float2)(0x1.2e0000p+0f, 0x1.2e025cp-15f),
- (float2)(0x1.2c0000p+0f, 0x1.3f69b0p-9f),
- (float2)(0x1.2a0000p+0f, 0x1.404ad0p-8f),
- (float2)(0x1.280000p+0f, 0x1.e4129ep-8f),
- (float2)(0x1.280000p+0f, 0x1.160252p-9f),
- (float2)(0x1.260000p+0f, 0x1.350b88p-8f),
- (float2)(0x1.240000p+0f, 0x1.e22708p-8f),
- (float2)(0x1.240000p+0f, 0x1.24924ap-9f),
- (float2)(0x1.220000p+0f, 0x1.45678ap-8f),
- (float2)(0x1.200000p+0f, 0x1.fb7812p-8f),
- (float2)(0x1.200000p+0f, 0x1.68e18cp-9f),
- (float2)(0x1.1e0000p+0f, 0x1.7047dcp-8f),
- (float2)(0x1.1e0000p+0f, 0x1.779da0p-11f),
- (float2)(0x1.1c0000p+0f, 0x1.e0d5b4p-9f),
- (float2)(0x1.1a0000p+0f, 0x1.b4a404p-8f),
- (float2)(0x1.1a0000p+0f, 0x1.ee5846p-10f),
- (float2)(0x1.180000p+0f, 0x1.453808p-8f),
- (float2)(0x1.180000p+0f, 0x1.181182p-12f),
- (float2)(0x1.160000p+0f, 0x1.c0d128p-9f),
- (float2)(0x1.140000p+0f, 0x1.b1e5f8p-8f),
- (float2)(0x1.140000p+0f, 0x1.0be1c2p-9f),
- (float2)(0x1.120000p+0f, 0x1.5c8114p-8f),
- (float2)(0x1.120000p+0f, 0x1.ac73aep-11f),
- (float2)(0x1.100000p+0f, 0x1.111112p-8f),
- (float2)(0x1.0e0000p+0f, 0x1.ef0110p-8f),
- (float2)(0x1.0e0000p+0f, 0x1.9ead7cp-9f),
- (float2)(0x1.0c0000p+0f, 0x1.b20a88p-8f),
- (float2)(0x1.0c0000p+0f, 0x1.2e29f8p-9f),
- (float2)(0x1.0a0000p+0f, 0x1.7e6ec2p-8f),
- (float2)(0x1.0a0000p+0f, 0x1.a0429ap-10f),
- (float2)(0x1.080000p+0f, 0x1.53f390p-8f),
- (float2)(0x1.080000p+0f, 0x1.084210p-10f),
- (float2)(0x1.060000p+0f, 0x1.3260a4p-8f),
- (float2)(0x1.060000p+0f, 0x1.26e978p-11f),
- (float2)(0x1.040000p+0f, 0x1.197f7ep-8f),
- (float2)(0x1.040000p+0f, 0x1.041042p-12f),
- (float2)(0x1.020000p+0f, 0x1.091b52p-8f),
- (float2)(0x1.020000p+0f, 0x1.020408p-14f),
- (float2)(0x1.000000p+0f, 0x1.010102p-8f),
- (float2)(0x1.000000p+0f, 0x0.000000p+0f),
-};
-
DECLARE_TABLE(float2, LOG2_TBL, 129) = {
(float2)(0x0.000000p+0f, 0x0.000000p+0f),
(float2)(0x1.6f8000p-7f, 0x1.942dbap-17f),
@@ -662,75 +530,6 @@ DECLARE_TABLE(float, EXP_TBL, 65) = {
0x1.000000p+1f,
};
-DECLARE_TABLE(float2, EXP_TBL_EP, 65) = {
- (float2) (0x1.000000p+0f, 0x0.000000p+0f),
- (float2) (0x1.02c000p+0f, 0x1.347ceep-13f),
- (float2) (0x1.058000p+0f, 0x1.b0d314p-12f),
- (float2) (0x1.084000p+0f, 0x1.a28c3ap-11f),
- (float2) (0x1.0b4000p+0f, 0x1.586cf8p-12f),
- (float2) (0x1.0e0000p+0f, 0x1.f61968p-11f),
- (float2) (0x1.110000p+0f, 0x1.80e808p-11f),
- (float2) (0x1.140000p+0f, 0x1.4d5754p-11f),
- (float2) (0x1.170000p+0f, 0x1.5c1e3ep-11f),
- (float2) (0x1.1a0000p+0f, 0x1.adf5b6p-11f),
- (float2) (0x1.1d4000p+0f, 0x1.0e62d0p-13f),
- (float2) (0x1.204000p+0f, 0x1.1dc430p-11f),
- (float2) (0x1.238000p+0f, 0x1.e9b9d4p-14f),
- (float2) (0x1.268000p+0f, 0x1.a2b2f0p-11f),
- (float2) (0x1.29c000p+0f, 0x1.4efa8ep-11f),
- (float2) (0x1.2d0000p+0f, 0x1.42d372p-11f),
- (float2) (0x1.304000p+0f, 0x1.7f0518p-11f),
- (float2) (0x1.33c000p+0f, 0x1.164c82p-17f),
- (float2) (0x1.370000p+0f, 0x1.a7373ap-12f),
- (float2) (0x1.3a4000p+0f, 0x1.ed9a72p-11f),
- (float2) (0x1.3dc000p+0f, 0x1.532608p-11f),
- (float2) (0x1.414000p+0f, 0x1.0510fap-11f),
- (float2) (0x1.44c000p+0f, 0x1.043030p-11f),
- (float2) (0x1.484000p+0f, 0x1.515ae0p-11f),
- (float2) (0x1.4bc000p+0f, 0x1.ed6a9ap-11f),
- (float2) (0x1.4f8000p+0f, 0x1.b2769cp-12f),
- (float2) (0x1.534000p+0f, 0x1.5ab4eap-15f),
- (float2) (0x1.56c000p+0f, 0x1.a39b5ap-11f),
- (float2) (0x1.5a8000p+0f, 0x1.83eea4p-11f),
- (float2) (0x1.5e4000p+0f, 0x1.b78ad6p-11f),
- (float2) (0x1.624000p+0f, 0x1.fac0e8p-14f),
- (float2) (0x1.660000p+0f, 0x1.1c412ap-11f),
- (float2) (0x1.6a0000p+0f, 0x1.3cccfep-13f),
- (float2) (0x1.6dc000p+0f, 0x1.d91e32p-11f),
- (float2) (0x1.71c000p+0f, 0x1.baf476p-11f),
- (float2) (0x1.75c000p+0f, 0x1.f5ab20p-11f),
- (float2) (0x1.7a0000p+0f, 0x1.1473eap-12f),
- (float2) (0x1.7e0000p+0f, 0x1.799b66p-11f),
- (float2) (0x1.824000p+0f, 0x1.89994cp-12f),
- (float2) (0x1.868000p+0f, 0x1.b33688p-13f),
- (float2) (0x1.8ac000p+0f, 0x1.ca8454p-13f),
- (float2) (0x1.8f0000p+0f, 0x1.ae9914p-12f),
- (float2) (0x1.934000p+0f, 0x1.9bd866p-11f),
- (float2) (0x1.97c000p+0f, 0x1.829fdep-12f),
- (float2) (0x1.9c4000p+0f, 0x1.230546p-13f),
- (float2) (0x1.a0c000p+0f, 0x1.99ed76p-14f),
- (float2) (0x1.a54000p+0f, 0x1.03b23ep-12f),
- (float2) (0x1.a9c000p+0f, 0x1.35aabcp-11f),
- (float2) (0x1.ae8000p+0f, 0x1.3f32b4p-13f),
- (float2) (0x1.b30000p+0f, 0x1.d15c26p-11f),
- (float2) (0x1.b7c000p+0f, 0x1.bb797cp-11f),
- (float2) (0x1.bcc000p+0f, 0x1.e904bcp-16f),
- (float2) (0x1.c18000p+0f, 0x1.9bdd84p-12f),
- (float2) (0x1.c64000p+0f, 0x1.f8972ap-11f),
- (float2) (0x1.cb4000p+0f, 0x1.906e76p-11f),
- (float2) (0x1.d04000p+0f, 0x1.96a502p-11f),
- (float2) (0x1.d58000p+0f, 0x1.8dcfbap-16f),
- (float2) (0x1.da8000p+0f, 0x1.e603dap-12f),
- (float2) (0x1.dfc000p+0f, 0x1.2e66f6p-13f),
- (float2) (0x1.e50000p+0f, 0x1.773c58p-15f),
- (float2) (0x1.ea4000p+0f, 0x1.5f4548p-13f),
- (float2) (0x1.ef8000p+0f, 0x1.0df730p-11f),
- (float2) (0x1.f50000p+0f, 0x1.d96db8p-14f),
- (float2) (0x1.fa4000p+0f, 0x1.e0c0cep-11f),
- (float2) (0x1.000000p+1f, 0x0.000000p+0f),
-};
-
-TABLE_FUNCTION(float2, LOG_INV_TBL_EP, log_inv_tbl_ep);
TABLE_FUNCTION(float2, LOG2_TBL, log2_tbl);
TABLE_FUNCTION(float2, LOG10_TBL, log10_tbl);
@@ -741,79 +540,9 @@ uint4 TABLE_MANGLE(pibits_tbl)(size_t idx) {
TABLE_FUNCTION(float2, SINHCOSH_TBL, sinhcosh_tbl);
TABLE_FUNCTION(float2, CBRT_TBL, cbrt_tbl);
TABLE_FUNCTION(float, EXP_TBL, exp_tbl);
-TABLE_FUNCTION(float2, EXP_TBL_EP, exp_tbl_ep);
#ifdef cl_khr_fp64
-DECLARE_TABLE(double2, TWO_TO_JBY64_EP, 64) = {
- (double2)(0x1.0000000000000p+0, 0x0.0000000000000p+0),
- (double2)(0x1.02c9a30000000p+0, 0x1.cef00c1dcdef9p-25),
- (double2)(0x1.059b0d0000000p+0, 0x1.8ac2ba1d73e2ap-27),
- (double2)(0x1.0874510000000p+0, 0x1.0eb37901186bep-25),
- (double2)(0x1.0b55860000000p+0, 0x1.9f3121ec53172p-25),
- (double2)(0x1.0e3ec30000000p+0, 0x1.69e8d10103a17p-27),
- (double2)(0x1.11301d0000000p+0, 0x1.25b50a4ebbf1ap-32),
- (double2)(0x1.1429aa0000000p+0, 0x1.d525bbf668203p-25),
- (double2)(0x1.172b830000000p+0, 0x1.8faa2f5b9bef9p-25),
- (double2)(0x1.1a35be0000000p+0, 0x1.6df96ea796d31p-25),
- (double2)(0x1.1d48730000000p+0, 0x1.68b9aa7805b80p-28),
- (double2)(0x1.2063b80000000p+0, 0x1.0c519ac771dd6p-25),
- (double2)(0x1.2387a60000000p+0, 0x1.ceac470cd83f5p-25),
- (double2)(0x1.26b4560000000p+0, 0x1.789f37495e99cp-26),
- (double2)(0x1.29e9df0000000p+0, 0x1.47f7b84b09745p-26),
- (double2)(0x1.2d285a0000000p+0, 0x1.b900c2d002475p-26),
- (double2)(0x1.306fe00000000p+0, 0x1.4636e2a5bd1abp-25),
- (double2)(0x1.33c08b0000000p+0, 0x1.320b7fa64e430p-27),
- (double2)(0x1.371a730000000p+0, 0x1.ceaa72a9c5154p-26),
- (double2)(0x1.3a7db30000000p+0, 0x1.3967fdba86f24p-26),
- (double2)(0x1.3dea640000000p+0, 0x1.82468446b6824p-25),
- (double2)(0x1.4160a20000000p+0, 0x1.f72e29f84325bp-28),
- (double2)(0x1.44e0860000000p+0, 0x1.8624b40c4dbd0p-30),
- (double2)(0x1.486a2b0000000p+0, 0x1.704f3404f068ep-26),
- (double2)(0x1.4bfdad0000000p+0, 0x1.4d8a89c750e5ep-26),
- (double2)(0x1.4f9b270000000p+0, 0x1.a74b29ab4cf62p-26),
- (double2)(0x1.5342b50000000p+0, 0x1.a753e077c2a0fp-26),
- (double2)(0x1.56f4730000000p+0, 0x1.ad49f699bb2c0p-26),
- (double2)(0x1.5ab07d0000000p+0, 0x1.a90a852b19260p-25),
- (double2)(0x1.5e76f10000000p+0, 0x1.6b48521ba6f93p-26),
- (double2)(0x1.6247eb0000000p+0, 0x1.d2ac258f87d03p-31),
- (double2)(0x1.6623880000000p+0, 0x1.2a91124893ecfp-27),
- (double2)(0x1.6a09e60000000p+0, 0x1.9fcef32422cbep-26),
- (double2)(0x1.6dfb230000000p+0, 0x1.8ca345de441c5p-25),
- (double2)(0x1.71f75e0000000p+0, 0x1.1d8bee7ba46e1p-25),
- (double2)(0x1.75feb50000000p+0, 0x1.9099f22fdba6ap-26),
- (double2)(0x1.7a11470000000p+0, 0x1.f580c36bea881p-27),
- (double2)(0x1.7e2f330000000p+0, 0x1.b3d398841740ap-26),
- (double2)(0x1.8258990000000p+0, 0x1.2999c25159f11p-25),
- (double2)(0x1.868d990000000p+0, 0x1.68925d901c83bp-25),
- (double2)(0x1.8ace540000000p+0, 0x1.15506dadd3e2ap-27),
- (double2)(0x1.8f1ae90000000p+0, 0x1.22aee6c57304ep-25),
- (double2)(0x1.93737b0000000p+0, 0x1.9b8bc9e8a0387p-29),
- (double2)(0x1.97d8290000000p+0, 0x1.fbc9c9f173d24p-25),
- (double2)(0x1.9c49180000000p+0, 0x1.51f8480e3e235p-27),
- (double2)(0x1.a0c6670000000p+0, 0x1.6bbcac96535b5p-25),
- (double2)(0x1.a5503b0000000p+0, 0x1.1f12ae45a1224p-27),
- (double2)(0x1.a9e6b50000000p+0, 0x1.5e7f6fd0fac90p-26),
- (double2)(0x1.ae89f90000000p+0, 0x1.2b5a75abd0e69p-25),
- (double2)(0x1.b33a2b0000000p+0, 0x1.09e2bf5ed7fa1p-25),
- (double2)(0x1.b7f76f0000000p+0, 0x1.7daf237553d84p-27),
- (double2)(0x1.bcc1e90000000p+0, 0x1.2f074891ee83dp-30),
- (double2)(0x1.c199bd0000000p+0, 0x1.b0aa538444196p-25),
- (double2)(0x1.c67f120000000p+0, 0x1.cafa29694426fp-25),
- (double2)(0x1.cb720d0000000p+0, 0x1.9df20d22a0797p-25),
- (double2)(0x1.d072d40000000p+0, 0x1.40f12f71a1e45p-25),
- (double2)(0x1.d5818d0000000p+0, 0x1.9f7490e4bb40bp-25),
- (double2)(0x1.da9e600000000p+0, 0x1.ed9942b84600dp-27),
- (double2)(0x1.dfc9730000000p+0, 0x1.bdcdaf5cb4656p-27),
- (double2)(0x1.e502ee0000000p+0, 0x1.e2cffd89cf44cp-26),
- (double2)(0x1.ea4afa0000000p+0, 0x1.52486cc2c7b9dp-27),
- (double2)(0x1.efa1be0000000p+0, 0x1.cc2b44eee3fa4p-25),
- (double2)(0x1.f507650000000p+0, 0x1.6dc8a80ce9f09p-25),
- (double2)(0x1.fa7c180000000p+0, 0x1.9e90d82e90a7ep-28)
-
-};
-
-
DECLARE_TABLE(double2, SINH_TBL, 37) = {
(double2)(0x0.0000000000000p+0, 0x0.0000000000000p+0),
(double2)(0x1.2cd9fc0000000p+0, 0x1.13ae6096a0092p-26),
@@ -1424,534 +1153,10 @@ DECLARE_TABLE(double2, CBRT_REM_TBL, 5) = {
};
-DECLARE_TABLE(double2, POWLOG_TBL, 258) = {
- (double2)(0x0.0000000000000p+0, 0x0.0000000000000p+0),
- (double2)(0x1.ff00aa0000000p-9, 0x1.5885e0250435ap-36),
- (double2)(0x1.fe02a60000000p-8, 0x1.620cf11f86ed2p-33),
- (double2)(0x1.7dc4750000000p-7, 0x1.f0214edba4a25p-32),
- (double2)(0x1.fc0a8b0000000p-7, 0x1.f807c79f3db4ep-36),
- (double2)(0x1.3cea440000000p-6, 0x1.a352ba779a52bp-33),
- (double2)(0x1.7b91b00000000p-6, 0x1.f56c46aa49fd5p-32),
- (double2)(0x1.b9fc020000000p-6, 0x1.ebe465fef5196p-32),
- (double2)(0x1.f829b00000000p-6, 0x1.cf0660099f1f8p-31),
- (double2)(0x1.1b0d980000000p-5, 0x1.247b2ff85945dp-30),
- (double2)(0x1.39e87b0000000p-5, 0x1.3fd7abf5202b6p-30),
- (double2)(0x1.58a5ba0000000p-5, 0x1.f91c9a918d51ep-30),
- (double2)(0x1.77458f0000000p-5, 0x1.8cb73f118d3cap-31),
- (double2)(0x1.95c8300000000p-5, 0x1.d91c7d6fad074p-30),
- (double2)(0x1.b42dd70000000p-5, 0x1.1971bec28d14cp-33),
- (double2)(0x1.d276b80000000p-5, 0x1.5b616a423c78ap-30),
- (double2)(0x1.f0a30c0000000p-5, 0x1.162a6617cc971p-37),
- (double2)(0x1.0759830000000p-4, 0x1.66391c4c06d29p-30),
- (double2)(0x1.16536e0000000p-4, 0x1.d46f5c1d0c4b8p-29),
- (double2)(0x1.253f620000000p-4, 0x1.e14282df1f6d3p-29),
- (double2)(0x1.341d790000000p-4, 0x1.86f47424a660dp-30),
- (double2)(0x1.42edcb0000000p-4, 0x1.d4c8de077753ep-29),
- (double2)(0x1.51b0730000000p-4, 0x1.e0c307ed24f1cp-29),
- (double2)(0x1.60658a0000000p-4, 0x1.26ea18763bdd3p-29),
- (double2)(0x1.6f0d280000000p-4, 0x1.5cad69737c933p-29),
- (double2)(0x1.7da7660000000p-4, 0x1.af62599088901p-29),
- (double2)(0x1.8c345d0000000p-4, 0x1.8c66c83d6b2d0p-30),
- (double2)(0x1.9ab4240000000p-4, 0x1.880ceb36fb30fp-30),
- (double2)(0x1.a926d30000000p-4, 0x1.495aac6ca17a4p-29),
- (double2)(0x1.b78c820000000p-4, 0x1.761db4210878cp-29),
- (double2)(0x1.c5e5480000000p-4, 0x1.eb78e862bac2fp-29),
- (double2)(0x1.d4313d0000000p-4, 0x1.9b2cd75790dd9p-30),
- (double2)(0x1.e270760000000p-4, 0x1.c55e5cbd3d50fp-29),
- (double2)(0x1.f0a30c0000000p-4, 0x1.162a6617cc971p-36),
- (double2)(0x1.fec9130000000p-4, 0x1.dbeabaaa2e519p-32),
- (double2)(0x1.0671510000000p-3, 0x1.652cb7150c647p-30),
- (double2)(0x1.0d77e70000000p-3, 0x1.9a11cb2cd2ee2p-28),
- (double2)(0x1.1478580000000p-3, 0x1.19d0ab1a28813p-29),
- (double2)(0x1.1b72ad0000000p-3, 0x1.4bd9e80a41811p-29),
- (double2)(0x1.2266f10000000p-3, 0x1.214b596faa3dfp-28),
- (double2)(0x1.29552f0000000p-3, 0x1.03fea46980bb8p-28),
- (double2)(0x1.303d710000000p-3, 0x1.1c8ffa5fd28c7p-28),
- (double2)(0x1.371fc20000000p-3, 0x1.e8f743bcd96c5p-35),
- (double2)(0x1.3dfc2b0000000p-3, 0x1.d98c5395315c6p-32),
- (double2)(0x1.44d2b60000000p-3, 0x1.996fa3ccfa7b2p-28),
- (double2)(0x1.4ba36f0000000p-3, 0x1.cd2af2ad13037p-30),
- (double2)(0x1.526e5e0000000p-3, 0x1.d0da1bd17200ep-30),
- (double2)(0x1.59338d0000000p-3, 0x1.330410ba68b75p-28),
- (double2)(0x1.5ff3070000000p-3, 0x1.4f27a790e7c41p-32),
- (double2)(0x1.66acd40000000p-3, 0x1.3956a86f6ff1bp-30),
- (double2)(0x1.6d60fe0000000p-3, 0x1.c6748723551d9p-29),
- (double2)(0x1.740f8f0000000p-3, 0x1.500de9326cdfcp-29),
- (double2)(0x1.7ab8900000000p-3, 0x1.086c848df1b59p-30),
- (double2)(0x1.815c0a0000000p-3, 0x1.4357ead6836ffp-31),
- (double2)(0x1.87fa060000000p-3, 0x1.4832442408024p-29),
- (double2)(0x1.8e928d0000000p-3, 0x1.d10da8154b13dp-28),
- (double2)(0x1.9525a90000000p-3, 0x1.9e8ad68ec8260p-28),
- (double2)(0x1.9bb3620000000p-3, 0x1.cfbf706abaf18p-28),
- (double2)(0x1.a23bc10000000p-3, 0x1.fc56ac6326e23p-28),
- (double2)(0x1.a8becf0000000p-3, 0x1.9105e3185cf21p-28),
- (double2)(0x1.af3c940000000p-3, 0x1.d017fe5b19cc0p-28),
- (double2)(0x1.b5b5190000000p-3, 0x1.d1f6b48dd13fep-28),
- (double2)(0x1.bc28670000000p-3, 0x1.0b63358a7e73ap-29),
- (double2)(0x1.c296850000000p-3, 0x1.63063028c211cp-29),
- (double2)(0x1.c8ff7c0000000p-3, 0x1.e6a6886b09760p-29),
- (double2)(0x1.cf63540000000p-3, 0x1.c138bb891cd03p-28),
- (double2)(0x1.d5c2160000000p-3, 0x1.69f7722b7221ap-28),
- (double2)(0x1.dc1bca0000000p-3, 0x1.57d8fac1a628cp-32),
- (double2)(0x1.e270760000000p-3, 0x1.c55e5cbd3d50fp-28),
- (double2)(0x1.e8c0250000000p-3, 0x1.552d2ff48fe2ep-30),
- (double2)(0x1.ef0adc0000000p-3, 0x1.7b8b26ca431bcp-28),
- (double2)(0x1.f550a50000000p-3, 0x1.92decdc1c5f6dp-29),
- (double2)(0x1.fb91860000000p-3, 0x1.abc7c551aaa8cp-28),
- (double2)(0x1.00e6c40000000p-2, 0x1.6b540731a354bp-28),
- (double2)(0x1.0402590000000p-2, 0x1.2d341036b89efp-28),
- (double2)(0x1.071b850000000p-2, 0x1.f9ab21a3a2e0fp-27),
- (double2)(0x1.0a324e0000000p-2, 0x1.39c871afb9fbdp-29),
- (double2)(0x1.0d46b50000000p-2, 0x1.e6add2c81f640p-28),
- (double2)(0x1.1058bf0000000p-2, 0x1.35c95aa313f41p-27),
- (double2)(0x1.1368700000000p-2, 0x1.49d4582f6cc53p-29),
- (double2)(0x1.1675ca0000000p-2, 0x1.7574c1c07398fp-27),
- (double2)(0x1.1980d20000000p-2, 0x1.ba846dece9e8dp-27),
- (double2)(0x1.1c898c0000000p-2, 0x1.6999fafbc68e7p-30),
- (double2)(0x1.1f8ff90000000p-2, 0x1.c9145e51b0103p-27),
- (double2)(0x1.22941f0000000p-2, 0x1.79ef2cb44850ap-27),
- (double2)(0x1.2596010000000p-2, 0x1.beec73de11275p-31),
- (double2)(0x1.2895a10000000p-2, 0x1.ef4351af5a498p-29),
- (double2)(0x1.2b93030000000p-2, 0x1.5713a493b4a50p-27),
- (double2)(0x1.2e8e2b0000000p-2, 0x1.5c23a61385992p-27),
- (double2)(0x1.31871c0000000p-2, 0x1.2a88309f57299p-27),
- (double2)(0x1.347dd90000000p-2, 0x1.530faa9ac8acep-27),
- (double2)(0x1.3772660000000p-2, 0x1.5fec2d792a758p-29),
- (double2)(0x1.3a64c50000000p-2, 0x1.5a517a71cbcd7p-28),
- (double2)(0x1.3d54fa0000000p-2, 0x1.707dc3e1cd9a3p-28),
- (double2)(0x1.4043080000000p-2, 0x1.a1a9f8ef43049p-28),
- (double2)(0x1.432ef20000000p-2, 0x1.409d0276b3674p-27),
- (double2)(0x1.4618bc0000000p-2, 0x1.0e2f613e85bd9p-29),
- (double2)(0x1.4900680000000p-2, 0x1.0027433001e5fp-32),
- (double2)(0x1.4be5f90000000p-2, 0x1.5dde2836d3265p-28),
- (double2)(0x1.4ec9730000000p-2, 0x1.300134d7aaf04p-29),
- (double2)(0x1.51aad80000000p-2, 0x1.cb7e0b42724f5p-28),
- (double2)(0x1.548a2c0000000p-2, 0x1.d6e93167e6308p-29),
- (double2)(0x1.5767710000000p-2, 0x1.d1569b1526adbp-28),
- (double2)(0x1.5a42ab0000000p-2, 0x1.e99fc338a1a41p-31),
- (double2)(0x1.5d1bdb0000000p-2, 0x1.eb01394a11b1cp-27),
- (double2)(0x1.5ff3070000000p-2, 0x1.4f27a790e7c41p-31),
- (double2)(0x1.62c82f0000000p-2, 0x1.5ce3ca97b7af9p-29),
- (double2)(0x1.659b570000000p-2, 0x1.81f0f940ed857p-29),
- (double2)(0x1.686c810000000p-2, 0x1.d36295d88857cp-27),
- (double2)(0x1.6b3bb20000000p-2, 0x1.1aca1ec4af526p-29),
- (double2)(0x1.6e08ea0000000p-2, 0x1.45743c7182726p-27),
- (double2)(0x1.70d42e0000000p-2, 0x1.3c491aead337ep-29),
- (double2)(0x1.739d7f0000000p-2, 0x1.aef401a738931p-28),
- (double2)(0x1.7664e10000000p-2, 0x1.1cede76092a29p-29),
- (double2)(0x1.792a550000000p-2, 0x1.fba8f44f82bb4p-27),
- (double2)(0x1.7bede00000000p-2, 0x1.46f5f7f3c3e1ap-27),
- (double2)(0x1.7eaf830000000p-2, 0x1.7055f86c9674bp-27),
- (double2)(0x1.816f410000000p-2, 0x1.b41a92b6b6e1ap-27),
- (double2)(0x1.842d1d0000000p-2, 0x1.43d162e927628p-27),
- (double2)(0x1.86e9190000000p-2, 0x1.466174013f9b1p-27),
- (double2)(0x1.89a3380000000p-2, 0x1.b05096ad69c62p-28),
- (double2)(0x1.8c5b7c0000000p-2, 0x1.0b169150faa58p-27),
- (double2)(0x1.8f11e80000000p-2, 0x1.cd98b1df85da7p-28),
- (double2)(0x1.91c67e0000000p-2, 0x1.68b507b0f8fa8p-27),
- (double2)(0x1.9479410000000p-2, 0x1.8422df57499bap-27),
- (double2)(0x1.972a340000000p-2, 0x1.1351586970274p-30),
- (double2)(0x1.99d9580000000p-2, 0x1.17e08acba92eep-30),
- (double2)(0x1.9c86b00000000p-2, 0x1.6e04314dd0229p-29),
- (double2)(0x1.9f323e0000000p-2, 0x1.97f3097e56d1ap-27),
- (double2)(0x1.a1dc060000000p-2, 0x1.356e655901286p-28),
- (double2)(0x1.a484090000000p-2, 0x1.cb761457f94d6p-31),
- (double2)(0x1.a72a490000000p-2, 0x1.9af67a85a9dacp-28),
- (double2)(0x1.a9cec90000000p-2, 0x1.53410931a909fp-27),
- (double2)(0x1.ac718c0000000p-2, 0x1.2c587206058f5p-29),
- (double2)(0x1.af12930000000p-2, 0x1.23bc358899c22p-29),
- (double2)(0x1.b1b1e00000000p-2, 0x1.d7bf8b6d223cbp-27),
- (double2)(0x1.b44f770000000p-2, 0x1.7991ec5197ddbp-27),
- (double2)(0x1.b6eb590000000p-2, 0x1.a79e6bb3a9219p-27),
- (double2)(0x1.b985890000000p-2, 0x1.a4c43ed663ec5p-28),
- (double2)(0x1.bc1e080000000p-2, 0x1.61b5a1484f438p-27),
- (double2)(0x1.beb4d90000000p-2, 0x1.b4e36f7ef0c3ap-27),
- (double2)(0x1.c149ff0000000p-2, 0x1.15f026acd0d1bp-30),
- (double2)(0x1.c3dd7a0000000p-2, 0x1.f36b535cecf05p-28),
- (double2)(0x1.c66f4e0000000p-2, 0x1.ffb7fbf3eb5c6p-29),
- (double2)(0x1.c8ff7c0000000p-2, 0x1.e6a6886b09760p-28),
- (double2)(0x1.cb8e070000000p-2, 0x1.135eb27f5bbc3p-28),
- (double2)(0x1.ce1af00000000p-2, 0x1.70be7d6f6fa57p-27),
- (double2)(0x1.d0a63a0000000p-2, 0x1.ce43cc84ab338p-27),
- (double2)(0x1.d32fe70000000p-2, 0x1.c01d7aac3bd91p-27),
- (double2)(0x1.d5b7f90000000p-2, 0x1.5c58d07961060p-27),
- (double2)(0x1.d83e720000000p-2, 0x1.628bcf941456ep-28),
- (double2)(0x1.dac3530000000p-2, 0x1.c58b2a8461cd2p-27),
- (double2)(0x1.dd46a00000000p-2, 0x1.3071282fb989ap-28),
- (double2)(0x1.dfc8590000000p-2, 0x1.20dab6a80f09cp-27),
- (double2)(0x1.e248810000000p-2, 0x1.4f8d84c397b1ep-27),
- (double2)(0x1.e4c71a0000000p-2, 0x1.0d0ee08599e48p-27),
- (double2)(0x1.e744260000000p-2, 0x1.d68787e37da36p-30),
- (double2)(0x1.e9bfa60000000p-2, 0x1.66187d591bafcp-28),
- (double2)(0x1.ec399d0000000p-2, 0x1.2346600bae772p-29),
- (double2)(0x1.eeb20c0000000p-2, 0x1.90377d0d61b8ep-28),
- (double2)(0x1.f128f50000000p-2, 0x1.f5e0dd966b907p-27),
- (double2)(0x1.f39e5b0000000p-2, 0x1.9023cb79a00e2p-27),
- (double2)(0x1.f6123f0000000p-2, 0x1.4e05158c28ad8p-27),
- (double2)(0x1.f884a30000000p-2, 0x1.bfa7b08b18ae4p-28),
- (double2)(0x1.faf5880000000p-2, 0x1.ef1e63db35f67p-27),
- (double2)(0x1.fd64f20000000p-2, 0x1.ec2ae39493d4fp-31),
- (double2)(0x1.ffd2e00000000p-2, 0x1.0afe930ab2fa0p-27),
- (double2)(0x1.011fab0000000p-1, 0x1.25ff8a1810dd4p-29),
- (double2)(0x1.02552a0000000p-1, 0x1.69743fb1a71a5p-27),
- (double2)(0x1.0389ee0000000p-1, 0x1.f9cc676785571p-26),
- (double2)(0x1.04bdf90000000p-1, 0x1.b524da4cbf982p-26),
- (double2)(0x1.05f14b0000000p-1, 0x1.a4c8b381535b8p-26),
- (double2)(0x1.0723e50000000p-1, 0x1.839be809caf2cp-26),
- (double2)(0x1.0855c80000000p-1, 0x1.0968a1cb82c13p-26),
- (double2)(0x1.0986f40000000p-1, 0x1.eae6a41723fb5p-26),
- (double2)(0x1.0ab76b0000000p-1, 0x1.d9c29a380a4dbp-26),
- (double2)(0x1.0be72e0000000p-1, 0x1.094aa0ada625ep-27),
- (double2)(0x1.0d163c0000000p-1, 0x1.973ad6fc108cap-26),
- (double2)(0x1.0e44980000000p-1, 0x1.747322fdbab97p-27),
- (double2)(0x1.0f72410000000p-1, 0x1.93692fa9d4221p-26),
- (double2)(0x1.109f390000000p-1, 0x1.c5a992dfbc7d9p-26),
- (double2)(0x1.11cb810000000p-1, 0x1.e1f33e102387ap-27),
- (double2)(0x1.12f7190000000p-1, 0x1.64fbef14c048cp-27),
- (double2)(0x1.1422020000000p-1, 0x1.490f513ca5e3bp-27),
- (double2)(0x1.154c3d0000000p-1, 0x1.7a6af4d4c799dp-28),
- (double2)(0x1.1675ca0000000p-1, 0x1.7574c1c07398fp-26),
- (double2)(0x1.179eab0000000p-1, 0x1.7b133417f8c1cp-26),
- (double2)(0x1.18c6e00000000p-1, 0x1.feb9e0c176514p-26),
- (double2)(0x1.19ee6b0000000p-1, 0x1.19f25bb3172f7p-27),
- (double2)(0x1.1b154b0000000p-1, 0x1.5f68a7bbfb852p-27),
- (double2)(0x1.1c3b810000000p-1, 0x1.ee278497929f1p-26),
- (double2)(0x1.1d610f0000000p-1, 0x1.ccee006109d58p-26),
- (double2)(0x1.1e85f50000000p-1, 0x1.ce081a07bd8b3p-26),
- (double2)(0x1.1faa340000000p-1, 0x1.70e12981817b8p-26),
- (double2)(0x1.20cdcd0000000p-1, 0x1.92ab6d93503d0p-29),
- (double2)(0x1.21f0bf0000000p-1, 0x1.8cb7dd7c3b61ep-26),
- (double2)(0x1.23130d0000000p-1, 0x1.efafd0a0b78dap-27),
- (double2)(0x1.2434b60000000p-1, 0x1.e907267c4288ep-26),
- (double2)(0x1.2555bc0000000p-1, 0x1.d31ef96780875p-26),
- (double2)(0x1.2676200000000p-1, 0x1.3430dfcd2ad50p-29),
- (double2)(0x1.2795e10000000p-1, 0x1.44d88d75bc1f9p-28),
- (double2)(0x1.28b5000000000p-1, 0x1.bec0f055e04fcp-26),
- (double2)(0x1.29d37f0000000p-1, 0x1.d85611590b9adp-26),
- (double2)(0x1.2af15f0000000p-1, 0x1.320568e583229p-32),
- (double2)(0x1.2c0e9e0000000p-1, 0x1.a891d1772f538p-26),
- (double2)(0x1.2d2b400000000p-1, 0x1.2edc9dabba74dp-29),
- (double2)(0x1.2e47430000000p-1, 0x1.b9009a1015086p-27),
- (double2)(0x1.2f62a90000000p-1, 0x1.2a12a8c5b1a19p-26),
- (double2)(0x1.307d730000000p-1, 0x1.a7885f0fdac85p-28),
- (double2)(0x1.3197a00000000p-1, 0x1.f4ffcd43ac691p-26),
- (double2)(0x1.32b1330000000p-1, 0x1.2243ae2640aadp-26),
- (double2)(0x1.33ca2b0000000p-1, 0x1.46513299035d3p-26),
- (double2)(0x1.34e2890000000p-1, 0x1.b39c3a62dd725p-26),
- (double2)(0x1.35fa4e0000000p-1, 0x1.ba6dd40049f51p-26),
- (double2)(0x1.37117b0000000p-1, 0x1.51d1ed7177409p-27),
- (double2)(0x1.38280f0000000p-1, 0x1.cb0f2fd7f5216p-26),
- (double2)(0x1.393e0d0000000p-1, 0x1.ab150cd4e2213p-28),
- (double2)(0x1.3a53730000000p-1, 0x1.cfd7bf3193844p-26),
- (double2)(0x1.3b68440000000p-1, 0x1.3fff8455f1dbdp-26),
- (double2)(0x1.3c7c7f0000000p-1, 0x1.fee640b905fc9p-26),
- (double2)(0x1.3d90260000000p-1, 0x1.4e2adf548084cp-26),
- (double2)(0x1.3ea3390000000p-1, 0x1.b597adc1ecdd2p-28),
- (double2)(0x1.3fb5b80000000p-1, 0x1.345bd096d3a75p-27),
- (double2)(0x1.40c7a40000000p-1, 0x1.101b9d2453c8bp-26),
- (double2)(0x1.41d8fe0000000p-1, 0x1.08ce55cc8c979p-26),
- (double2)(0x1.42e9c60000000p-1, 0x1.bbf017e595f71p-26),
- (double2)(0x1.43f9fe0000000p-1, 0x1.7ce733bd393dcp-28),
- (double2)(0x1.4509a50000000p-1, 0x1.33bb0a503f8a1p-29),
- (double2)(0x1.4618bc0000000p-1, 0x1.0e2f613e85bd9p-28),
- (double2)(0x1.4727430000000p-1, 0x1.e67555a635b3cp-26),
- (double2)(0x1.48353d0000000p-1, 0x1.ea88df73d5e8bp-29),
- (double2)(0x1.4942a80000000p-1, 0x1.d17e03bda18a8p-28),
- (double2)(0x1.4a4f850000000p-1, 0x1.b607d76044f7ep-26),
- (double2)(0x1.4b5bd60000000p-1, 0x1.2adc4e71bc2fcp-26),
- (double2)(0x1.4c679a0000000p-1, 0x1.f99dc7362d1d9p-26),
- (double2)(0x1.4d72d30000000p-1, 0x1.473fa008e6a6ap-26),
- (double2)(0x1.4e7d810000000p-1, 0x1.b75bb09cb0985p-29),
- (double2)(0x1.4f87a30000000p-1, 0x1.ea04dd10b9abap-26),
- (double2)(0x1.50913c0000000p-1, 0x1.802d0d6979674p-26),
- (double2)(0x1.519a4c0000000p-1, 0x1.74688ccd99094p-30),
- (double2)(0x1.52a2d20000000p-1, 0x1.96f16abb9df22p-27),
- (double2)(0x1.53aad00000000p-1, 0x1.6e66df2aa374fp-27),
- (double2)(0x1.54b2460000000p-1, 0x1.e66525ea4550ap-27),
- (double2)(0x1.55b9350000000p-1, 0x1.2d02f34f20cbdp-27),
- (double2)(0x1.56bf9d0000000p-1, 0x1.6cfce65047188p-27),
- (double2)(0x1.57c57f0000000p-1, 0x1.9b78c842d58b8p-28),
- (double2)(0x1.58cadb0000000p-1, 0x1.735e624c24bc9p-27),
- (double2)(0x1.59cfb20000000p-1, 0x1.7eba1f7dd1adfp-27),
- (double2)(0x1.5ad4040000000p-1, 0x1.86b3e59f65355p-26),
- (double2)(0x1.5bd7d30000000p-1, 0x1.ce38e637f1b4dp-30),
- (double2)(0x1.5cdb1d0000000p-1, 0x1.8d82ec919edc7p-26),
- (double2)(0x1.5ddde50000000p-1, 0x1.c52648ddcfa37p-27),
- (double2)(0x1.5ee02a0000000p-1, 0x1.2482ceae1ac12p-26),
- (double2)(0x1.5fe1ed0000000p-1, 0x1.5a312311aba4fp-26),
- (double2)(0x1.60e32f0000000p-1, 0x1.11e236329f225p-27),
- (double2)(0x1.61e3ef0000000p-1, 0x1.b48c8cd2f246cp-26),
- (double2)(0x1.62e42e0000000p-1, 0x1.efa39ef35793cp-25),
- (double2)(0x0.0000000000000p+0, 0x0.0000000000000p+0),
-};
-
-DECLARE_TABLE(double2, LOG_F_INV_TBL, 258) = {
- (double2)(0x1.0000000000000p+1, 0x0.0000000000000p+0),
- (double2)(0x1.fe00000000000p+0, 0x1.fe01fe01fe020p-16),
- (double2)(0x1.fc00000000000p+0, 0x1.fc07f01fc07f0p-14),
- (double2)(0x1.fa00000000000p+0, 0x1.1caa01fa11caap-12),
- (double2)(0x1.f800000000000p+0, 0x1.f81f81f81f820p-12),
- (double2)(0x1.f600000000000p+0, 0x1.8856506ddaba6p-11),
- (double2)(0x1.f400000000000p+0, 0x1.196792909c560p-10),
- (double2)(0x1.f200000000000p+0, 0x1.7d9108c2ad433p-10),
- (double2)(0x1.f000000000000p+0, 0x1.f07c1f07c1f08p-10),
- (double2)(0x1.ee00000000000p+0, 0x1.38ff08b1c03ddp-9),
- (double2)(0x1.ec00000000000p+0, 0x1.80f6603d980f6p-9),
- (double2)(0x1.ea00000000000p+0, 0x1.d00f57403d5d0p-9),
- (double2)(0x1.e900000000000p+0, 0x1.31abf0b7672a0p-12),
- (double2)(0x1.e700000000000p+0, 0x1.06a965d43919bp-10),
- (double2)(0x1.e500000000000p+0, 0x1.ceb240795ceb2p-10),
- (double2)(0x1.e300000000000p+0, 0x1.522f3b834e67fp-9),
- (double2)(0x1.e100000000000p+0, 0x1.c3c3c3c3c3c3cp-9),
- (double2)(0x1.e000000000000p+0, 0x1.e01e01e01e01ep-12),
- (double2)(0x1.de00000000000p+0, 0x1.75b8fe21a291cp-10),
- (double2)(0x1.dc00000000000p+0, 0x1.403b9403b9404p-9),
- (double2)(0x1.da00000000000p+0, 0x1.cc0ed7303b5ccp-9),
- (double2)(0x1.d900000000000p+0, 0x1.79118f3fc4da2p-11),
- (double2)(0x1.d700000000000p+0, 0x1.ed952e0b0ce46p-10),
- (double2)(0x1.d500000000000p+0, 0x1.95900eae56404p-9),
- (double2)(0x1.d400000000000p+0, 0x1.d41d41d41d41dp-12),
- (double2)(0x1.d200000000000p+0, 0x1.cb28ff16c69aep-10),
- (double2)(0x1.d000000000000p+0, 0x1.96b1edd80e866p-9),
- (double2)(0x1.cf00000000000p+0, 0x1.372e225fe30d9p-11),
- (double2)(0x1.cd00000000000p+0, 0x1.0ad12073615a2p-9),
- (double2)(0x1.cb00000000000p+0, 0x1.cdb2c0397cdb3p-9),
- (double2)(0x1.ca00000000000p+0, 0x1.2cc157b864407p-10),
- (double2)(0x1.c800000000000p+0, 0x1.64cb5f7148404p-9),
- (double2)(0x1.c700000000000p+0, 0x1.c71c71c71c71cp-12),
- (double2)(0x1.c500000000000p+0, 0x1.129a21a930b84p-9),
- (double2)(0x1.c300000000000p+0, 0x1.f1e0387f1e038p-9),
- (double2)(0x1.c200000000000p+0, 0x1.ad4e4ba80709bp-10),
- (double2)(0x1.c000000000000p+0, 0x1.c0e070381c0e0p-9),
- (double2)(0x1.bf00000000000p+0, 0x1.60fba1a362bb0p-10),
- (double2)(0x1.bd00000000000p+0, 0x1.a5713280dee96p-9),
- (double2)(0x1.bc00000000000p+0, 0x1.3f59620f9ece9p-10),
- (double2)(0x1.ba00000000000p+0, 0x1.9f22983759f23p-9),
- (double2)(0x1.b900000000000p+0, 0x1.478ac63fc8d5cp-10),
- (double2)(0x1.b700000000000p+0, 0x1.ad87bb4671656p-9),
- (double2)(0x1.b600000000000p+0, 0x1.78b8efbb8148cp-10),
- (double2)(0x1.b400000000000p+0, 0x1.d0369d0369d03p-9),
- (double2)(0x1.b300000000000p+0, 0x1.d212b601b3748p-10),
- (double2)(0x1.b200000000000p+0, 0x1.b2036406c80d9p-15),
- (double2)(0x1.b000000000000p+0, 0x1.29663b24547d1p-9),
- (double2)(0x1.af00000000000p+0, 0x1.435e50d79435ep-11),
- (double2)(0x1.ad00000000000p+0, 0x1.7d0ff2920bc03p-9),
- (double2)(0x1.ac00000000000p+0, 0x1.5c06b15c06b16p-10),
- (double2)(0x1.aa00000000000p+0, 0x1.e3a5f0fd7f954p-9),
- (double2)(0x1.a900000000000p+0, 0x1.1dec0d4c77b03p-9),
- (double2)(0x1.a800000000000p+0, 0x1.73289870ac52ep-11),
- (double2)(0x1.a600000000000p+0, 0x1.a034da034da03p-9),
- (double2)(0x1.a500000000000p+0, 0x1.d041da2292856p-10),
- (double2)(0x1.a400000000000p+0, 0x1.a41a41a41a41ap-12),
- (double2)(0x1.a200000000000p+0, 0x1.8550f8a39409dp-9),
- (double2)(0x1.a100000000000p+0, 0x1.b4fe5e92c0686p-10),
- (double2)(0x1.a000000000000p+0, 0x1.a01a01a01a01ap-12),
- (double2)(0x1.9e00000000000p+0, 0x1.91d2a2067b23ap-9),
- (double2)(0x1.9d00000000000p+0, 0x1.e7c5dada0b4e5p-10),
- (double2)(0x1.9c00000000000p+0, 0x1.68a7725080ce1p-11),
- (double2)(0x1.9a00000000000p+0, 0x1.c49d4aa21b490p-9),
- (double2)(0x1.9900000000000p+0, 0x1.3333333333333p-9),
- (double2)(0x1.9800000000000p+0, 0x1.4bc363b03fccfp-10),
- (double2)(0x1.9700000000000p+0, 0x1.c9f01970e4f81p-13),
- (double2)(0x1.9500000000000p+0, 0x1.97617c6ef5b25p-9),
- (double2)(0x1.9400000000000p+0, 0x1.161f9add3c0cap-9),
- (double2)(0x1.9300000000000p+0, 0x1.319fe6cb39806p-10),
- (double2)(0x1.9200000000000p+0, 0x1.f693a1c451ab3p-13),
- (double2)(0x1.9000000000000p+0, 0x1.a9e240321a9e2p-9),
- (double2)(0x1.8f00000000000p+0, 0x1.3831f3831f383p-9),
- (double2)(0x1.8e00000000000p+0, 0x1.949ebc4dcfc1cp-10),
- (double2)(0x1.8d00000000000p+0, 0x1.80c6980c6980cp-11),
- (double2)(0x1.8b00000000000p+0, 0x1.f9d00c5fe7403p-9),
- (double2)(0x1.8a00000000000p+0, 0x1.9721ed7e75347p-9),
- (double2)(0x1.8900000000000p+0, 0x1.381ec0313381fp-9),
- (double2)(0x1.8800000000000p+0, 0x1.b97c2aec12653p-10),
- (double2)(0x1.8700000000000p+0, 0x1.09ef3024ae3bap-10),
- (double2)(0x1.8600000000000p+0, 0x1.8618618618618p-12),
- (double2)(0x1.8400000000000p+0, 0x1.e0184f00c2780p-9),
- (double2)(0x1.8300000000000p+0, 0x1.92ef5657dba52p-9),
- (double2)(0x1.8200000000000p+0, 0x1.4940305494030p-9),
- (double2)(0x1.8100000000000p+0, 0x1.0303030303030p-9),
- (double2)(0x1.8000000000000p+0, 0x1.8060180601806p-10),
- (double2)(0x1.7f00000000000p+0, 0x1.017f405fd017fp-10),
- (double2)(0x1.7e00000000000p+0, 0x1.12a8ad278e8ddp-11),
- (double2)(0x1.7d00000000000p+0, 0x1.7d05f417d05f4p-14),
- (double2)(0x1.7b00000000000p+0, 0x1.d67245c02f7d6p-9),
- (double2)(0x1.7a00000000000p+0, 0x1.a4411c1d986a9p-9),
- (double2)(0x1.7900000000000p+0, 0x1.754d76c7316dfp-9),
- (double2)(0x1.7800000000000p+0, 0x1.49902f149902fp-9),
- (double2)(0x1.7700000000000p+0, 0x1.21023358c1a68p-9),
- (double2)(0x1.7600000000000p+0, 0x1.f7390d2a6c406p-10),
- (double2)(0x1.7500000000000p+0, 0x1.b2b0805d5b2b1p-10),
- (double2)(0x1.7400000000000p+0, 0x1.745d1745d1746p-10),
- (double2)(0x1.7300000000000p+0, 0x1.3c31507fa32c4p-10),
- (double2)(0x1.7200000000000p+0, 0x1.0a1fd1b7af017p-10),
- (double2)(0x1.7100000000000p+0, 0x1.bc36ce3e0453ap-11),
- (double2)(0x1.7000000000000p+0, 0x1.702e05c0b8170p-11),
- (double2)(0x1.6f00000000000p+0, 0x1.300b79300b793p-11),
- (double2)(0x1.6e00000000000p+0, 0x1.f76b4337c6cb1p-12),
- (double2)(0x1.6d00000000000p+0, 0x1.a62681c860fb0p-12),
- (double2)(0x1.6c00000000000p+0, 0x1.6c16c16c16c17p-12),
- (double2)(0x1.6b00000000000p+0, 0x1.490aa31a3cfc7p-12),
- (double2)(0x1.6a00000000000p+0, 0x1.3cd153729043ep-12),
- (double2)(0x1.6900000000000p+0, 0x1.473a88d0bfd2ep-12),
- (double2)(0x1.6800000000000p+0, 0x1.6816816816817p-12),
- (double2)(0x1.6700000000000p+0, 0x1.9f36016719f36p-12),
- (double2)(0x1.6600000000000p+0, 0x1.ec6a5122f9016p-12),
- (double2)(0x1.6500000000000p+0, 0x1.27c29da5519cfp-11),
- (double2)(0x1.6400000000000p+0, 0x1.642c8590b2164p-11),
- (double2)(0x1.6300000000000p+0, 0x1.ab5c45606f00bp-11),
- (double2)(0x1.6200000000000p+0, 0x1.fd3b80b11fd3cp-11),
- (double2)(0x1.6100000000000p+0, 0x1.2cda0c6ba4eaap-10),
- (double2)(0x1.6000000000000p+0, 0x1.6058160581606p-10),
- (double2)(0x1.5f00000000000p+0, 0x1.990d0a4b7ef87p-10),
- (double2)(0x1.5e00000000000p+0, 0x1.d6ee340579d6fp-10),
- (double2)(0x1.5d00000000000p+0, 0x1.0cf87d9c54a69p-9),
- (double2)(0x1.5c00000000000p+0, 0x1.310572620ae4cp-9),
- (double2)(0x1.5b00000000000p+0, 0x1.5798c8ff522a2p-9),
- (double2)(0x1.5a00000000000p+0, 0x1.80ad602b580adp-9),
- (double2)(0x1.5900000000000p+0, 0x1.ac3e24799546fp-9),
- (double2)(0x1.5800000000000p+0, 0x1.da46102b1da46p-9),
- (double2)(0x1.5800000000000p+0, 0x1.5805601580560p-14),
- (double2)(0x1.5700000000000p+0, 0x1.ed3c506b39a23p-12),
- (double2)(0x1.5600000000000p+0, 0x1.cbdd3e2970f60p-11),
- (double2)(0x1.5500000000000p+0, 0x1.5555555555555p-10),
- (double2)(0x1.5400000000000p+0, 0x1.c979aee0bf805p-10),
- (double2)(0x1.5300000000000p+0, 0x1.21291e81fd58ep-9),
- (double2)(0x1.5200000000000p+0, 0x1.5fead500a9580p-9),
- (double2)(0x1.5100000000000p+0, 0x1.a0fd5c5f02a3ap-9),
- (double2)(0x1.5000000000000p+0, 0x1.e45c223898adcp-9),
- (double2)(0x1.5000000000000p+0, 0x1.5015015015015p-12),
- (double2)(0x1.4f00000000000p+0, 0x1.c7b16ea64d422p-11),
- (double2)(0x1.4e00000000000p+0, 0x1.7829cbc14e5e1p-10),
- (double2)(0x1.4d00000000000p+0, 0x1.0877db8589720p-9),
- (double2)(0x1.4c00000000000p+0, 0x1.5710e4b5edceap-9),
- (double2)(0x1.4b00000000000p+0, 0x1.a7dbb4d1fc1c8p-9),
- (double2)(0x1.4a00000000000p+0, 0x1.fad40a57eb503p-9),
- (double2)(0x1.4a00000000000p+0, 0x1.3fd6bb00a5140p-11),
- (double2)(0x1.4900000000000p+0, 0x1.4e78ecb419ba9p-10),
- (double2)(0x1.4800000000000p+0, 0x1.00a44029100a4p-9),
- (double2)(0x1.4700000000000p+0, 0x1.5c28f5c28f5c3p-9),
- (double2)(0x1.4600000000000p+0, 0x1.b9c68b2c0cc4ap-9),
- (double2)(0x1.4600000000000p+0, 0x1.978feb9f34381p-13),
- (double2)(0x1.4500000000000p+0, 0x1.ecf163bb6500ap-11),
- (double2)(0x1.4400000000000p+0, 0x1.be1958b67ebb9p-10),
- (double2)(0x1.4300000000000p+0, 0x1.44e6157dc9a3bp-9),
- (double2)(0x1.4200000000000p+0, 0x1.acc4baa3f0ddfp-9),
- (double2)(0x1.4200000000000p+0, 0x1.6a4cbcb2a247bp-13),
- (double2)(0x1.4100000000000p+0, 0x1.0505050505050p-10),
- (double2)(0x1.4000000000000p+0, 0x1.e0b4439959819p-10),
- (double2)(0x1.3f00000000000p+0, 0x1.6027f6027f602p-9),
- (double2)(0x1.3e00000000000p+0, 0x1.d1e854b5e0db4p-9),
- (double2)(0x1.3e00000000000p+0, 0x1.165e7254813e2p-11),
- (double2)(0x1.3d00000000000p+0, 0x1.76646a9d716efp-10),
- (double2)(0x1.3c00000000000p+0, 0x1.32b48f757ce88p-9),
- (double2)(0x1.3b00000000000p+0, 0x1.ac1b24652a906p-9),
- (double2)(0x1.3b00000000000p+0, 0x1.3b13b13b13b14p-12),
- (double2)(0x1.3a00000000000p+0, 0x1.490e1eb208984p-10),
- (double2)(0x1.3900000000000p+0, 0x1.2385830fec66ep-9),
- (double2)(0x1.3800000000000p+0, 0x1.a45a6cc111b7ep-9),
- (double2)(0x1.3800000000000p+0, 0x1.3813813813814p-12),
- (double2)(0x1.3700000000000p+0, 0x1.56f472517b708p-10),
- (double2)(0x1.3600000000000p+0, 0x1.31be7bc0e8f2ap-9),
- (double2)(0x1.3500000000000p+0, 0x1.b9cbf3e55f044p-9),
- (double2)(0x1.3500000000000p+0, 0x1.0e7d95bc609a9p-11),
- (double2)(0x1.3400000000000p+0, 0x1.9e6b3804d19e7p-10),
- (double2)(0x1.3300000000000p+0, 0x1.5c8b6af7963c2p-9),
- (double2)(0x1.3200000000000p+0, 0x1.eb9dad43bf402p-9),
- (double2)(0x1.3200000000000p+0, 0x1.f1a515885fb37p-11),
- (double2)(0x1.3100000000000p+0, 0x1.0eeb1d3d76c02p-9),
- (double2)(0x1.3000000000000p+0, 0x1.a320261a32026p-9),
- (double2)(0x1.3000000000000p+0, 0x1.c82ac40260390p-12),
- (double2)(0x1.2f00000000000p+0, 0x1.a12f684bda12fp-10),
- (double2)(0x1.2e00000000000p+0, 0x1.69d43fda2962cp-9),
- (double2)(0x1.2e00000000000p+0, 0x1.2e025c04b8097p-15),
- (double2)(0x1.2d00000000000p+0, 0x1.42804b542804bp-10),
- (double2)(0x1.2c00000000000p+0, 0x1.3f69b02593f6ap-9),
- (double2)(0x1.2b00000000000p+0, 0x1.df31cb46e21fap-9),
- (double2)(0x1.2b00000000000p+0, 0x1.012b404ad012bp-10),
- (double2)(0x1.2a00000000000p+0, 0x1.23925e7820a7fp-9),
- (double2)(0x1.2900000000000p+0, 0x1.c8253c8253c82p-9),
- (double2)(0x1.2900000000000p+0, 0x1.b92ddc02526e5p-11),
- (double2)(0x1.2800000000000p+0, 0x1.1602511602511p-9),
- (double2)(0x1.2700000000000p+0, 0x1.bf471439c9adfp-9),
- (double2)(0x1.2700000000000p+0, 0x1.a85c40939a85cp-11),
- (double2)(0x1.2600000000000p+0, 0x1.166f9ac024d16p-9),
- (double2)(0x1.2500000000000p+0, 0x1.c44e10125e227p-9),
- (double2)(0x1.2500000000000p+0, 0x1.cebf48bbd90e5p-11),
- (double2)(0x1.2400000000000p+0, 0x1.2492492492492p-9),
- (double2)(0x1.2300000000000p+0, 0x1.d6f2e2ec0b673p-9),
- (double2)(0x1.2300000000000p+0, 0x1.159e26af37c05p-10),
- (double2)(0x1.2200000000000p+0, 0x1.4024540245402p-9),
- (double2)(0x1.2100000000000p+0, 0x1.f6f0243f6f024p-9),
- (double2)(0x1.2100000000000p+0, 0x1.5e60121579805p-10),
- (double2)(0x1.2000000000000p+0, 0x1.68e18cf81b10fp-9),
- (double2)(0x1.2000000000000p+0, 0x1.2012012012012p-12),
- (double2)(0x1.1f00000000000p+0, 0x1.c11f7047dc11fp-10),
- (double2)(0x1.1e00000000000p+0, 0x1.9e878ff70985ep-9),
- (double2)(0x1.1e00000000000p+0, 0x1.779d9fdc3a219p-11),
- (double2)(0x1.1d00000000000p+0, 0x1.1eace5c957907p-9),
- (double2)(0x1.1c00000000000p+0, 0x1.e0d5b450239e1p-9),
- (double2)(0x1.1c00000000000p+0, 0x1.48bf073816367p-10),
- (double2)(0x1.1b00000000000p+0, 0x1.694808dda5202p-9),
- (double2)(0x1.1b00000000000p+0, 0x1.7c67f2bae2b21p-12),
- (double2)(0x1.1a00000000000p+0, 0x1.ee58469ee5847p-10),
- (double2)(0x1.1900000000000p+0, 0x1.c0233c0233c02p-9),
- (double2)(0x1.1900000000000p+0, 0x1.14e02328a7012p-10),
- (double2)(0x1.1800000000000p+0, 0x1.561072057b573p-9),
- (double2)(0x1.1800000000000p+0, 0x1.1811811811812p-12),
- (double2)(0x1.1700000000000p+0, 0x1.e28646f5a1060p-10),
- (double2)(0x1.1600000000000p+0, 0x1.c0d1284e6f1d7p-9),
- (double2)(0x1.1600000000000p+0, 0x1.23543f0c80459p-10),
- (double2)(0x1.1500000000000p+0, 0x1.63cbeea4e1a09p-9),
- (double2)(0x1.1500000000000p+0, 0x1.b9a3fdd5c8cb8p-12),
- (double2)(0x1.1400000000000p+0, 0x1.0be1c159a76d2p-9),
- (double2)(0x1.1300000000000p+0, 0x1.e1d1a688e4838p-9),
- (double2)(0x1.1300000000000p+0, 0x1.72044d72044d7p-10),
- (double2)(0x1.1200000000000p+0, 0x1.91713db81577bp-9),
- (double2)(0x1.1200000000000p+0, 0x1.ac73ae9819b50p-11),
- (double2)(0x1.1100000000000p+0, 0x1.460334e904cf6p-9),
- (double2)(0x1.1100000000000p+0, 0x1.1111111111111p-12),
- (double2)(0x1.1000000000000p+0, 0x1.feef80441fef0p-10),
- (double2)(0x1.0f00000000000p+0, 0x1.de021fde021fep-9),
- (double2)(0x1.0f00000000000p+0, 0x1.7b7eacc9686a0p-10),
- (double2)(0x1.0e00000000000p+0, 0x1.9ead7cd391fbcp-9),
- (double2)(0x1.0e00000000000p+0, 0x1.0195609804390p-10),
- (double2)(0x1.0d00000000000p+0, 0x1.641511e8d2b32p-9),
- (double2)(0x1.0d00000000000p+0, 0x1.222b1acf1ce96p-11),
- (double2)(0x1.0c00000000000p+0, 0x1.2e29f79b47582p-9),
- (double2)(0x1.0c00000000000p+0, 0x1.4f0d1682e11cdp-13),
- (double2)(0x1.0b00000000000p+0, 0x1.f9bb096771e4dp-10),
- (double2)(0x1.0a00000000000p+0, 0x1.e5ee45dd96ae2p-9),
- (double2)(0x1.0a00000000000p+0, 0x1.a0429a0429a04p-10),
- (double2)(0x1.0900000000000p+0, 0x1.bb74d5f06c021p-9),
- (double2)(0x1.0900000000000p+0, 0x1.4fce404254fcep-10),
- (double2)(0x1.0800000000000p+0, 0x1.95766eacbc402p-9),
- (double2)(0x1.0800000000000p+0, 0x1.0842108421084p-10),
- (double2)(0x1.0700000000000p+0, 0x1.73e5371d5c338p-9),
- (double2)(0x1.0700000000000p+0, 0x1.930523fbe3368p-11),
- (double2)(0x1.0600000000000p+0, 0x1.56b38f225f6c4p-9),
- (double2)(0x1.0600000000000p+0, 0x1.26e978d4fdf3bp-11),
- (double2)(0x1.0500000000000p+0, 0x1.3dd40e4eb0cc6p-9),
- (double2)(0x1.0500000000000p+0, 0x1.97f7d73404146p-12),
- (double2)(0x1.0400000000000p+0, 0x1.293982cc98af1p-9),
- (double2)(0x1.0400000000000p+0, 0x1.0410410410410p-12),
- (double2)(0x1.0300000000000p+0, 0x1.18d6f048ff7e4p-9),
- (double2)(0x1.0300000000000p+0, 0x1.236a3ebc349dep-13),
- (double2)(0x1.0200000000000p+0, 0x1.0c9f8ee53d18cp-9),
- (double2)(0x1.0200000000000p+0, 0x1.0204081020408p-14),
- (double2)(0x1.0100000000000p+0, 0x1.0486ca2f46ea6p-9),
- (double2)(0x1.0100000000000p+0, 0x1.0101010101010p-16),
- (double2)(0x1.0000000000000p+0, 0x1.0080402010080p-9),
- (double2)(0x1.0000000000000p+0, 0x0.0000000000000p+0),
-};
-
-TABLE_FUNCTION(double2, TWO_TO_JBY64_EP, two_to_jby64_ep_tbl);
TABLE_FUNCTION(double2, SINH_TBL, sinh_tbl);
TABLE_FUNCTION(double2, COSH_TBL, cosh_tbl);
TABLE_FUNCTION(double, CBRT_INV_TBL, cbrt_inv_tbl);
TABLE_FUNCTION(double2, CBRT_DBL_TBL, cbrt_dbl_tbl);
TABLE_FUNCTION(double2, CBRT_REM_TBL, cbrt_rem_tbl);
-TABLE_FUNCTION(double2, POWLOG_TBL, powlog_tbl);
-TABLE_FUNCTION(double2, LOG_F_INV_TBL, log_f_inv_tbl);
#endif // cl_khr_fp64
diff --git a/libclc/spirv/lib/SOURCES b/libclc/spirv/lib/SOURCES
index 6be5107c5e736..921a257814817 100644
--- a/libclc/spirv/lib/SOURCES
+++ b/libclc/spirv/lib/SOURCES
@@ -58,11 +58,8 @@ math/fma.cl
../../generic/lib/math/logb.cl
../../generic/lib/math/modf.cl
../../generic/lib/math/tables.cl
-../../generic/lib/math/clc_pow.cl
../../generic/lib/math/pow.cl
-../../generic/lib/math/clc_pown.cl
../../generic/lib/math/pown.cl
-../../generic/lib/math/clc_powr.cl
../../generic/lib/math/powr.cl
../../generic/lib/math/remainder.cl
../../generic/lib/math/remquo.cl
More information about the cfe-commits
mailing list