[libclc] Move sinh, cosh & tanh to the CLC library

libclc/clc/include/clc/math/clc_cosh.h

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_MATH_CLC_COSH_H__
+#define __CLC_MATH_CLC_COSH_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_cosh
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_COSH_H__

libclc/clc/include/clc/math/clc_sinh.h

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_MATH_CLC_SINH_H__
+#define __CLC_MATH_CLC_SINH_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_sinh
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_SINH_H__

libclc/clc/include/clc/math/clc_tanh.h

@@ -0,0 +1,20 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef __CLC_MATH_CLC_TANH_H__
+#define __CLC_MATH_CLC_TANH_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_tanh
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_TANH_H__

libclc/clc/include/clc/math/tables.h

@@ -62,7 +62,6 @@
 TABLE_FUNCTION_DECL(float2, log2_tbl);
 TABLE_FUNCTION_DECL(float2, log10_tbl);
 TABLE_FUNCTION_DECL(uint4, pibits_tbl);
-TABLE_FUNCTION_DECL(float2, sinhcosh_tbl);
 
 CLC_TABLE_FUNCTION_DECL(float, log_inv_tbl_ep_head);
 CLC_TABLE_FUNCTION_DECL(float, log_inv_tbl_ep_tail);
@@ -74,6 +73,8 @@ CLC_TABLE_FUNCTION_DECL(float, exp_tbl_ep_head);
 CLC_TABLE_FUNCTION_DECL(float, exp_tbl_ep_tail);
 CLC_TABLE_FUNCTION_DECL(float, cbrt_tbl_head);
 CLC_TABLE_FUNCTION_DECL(float, cbrt_tbl_tail);
+CLC_TABLE_FUNCTION_DECL(float, sinhcosh_tbl_head);
+CLC_TABLE_FUNCTION_DECL(float, sinhcosh_tbl_tail);
 
 #ifdef cl_khr_fp64
 
@@ -85,8 +86,10 @@ CLC_TABLE_FUNCTION_DECL(double, atan_jby256_tbl_head);
 CLC_TABLE_FUNCTION_DECL(double, atan_jby256_tbl_tail);
 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);
+CLC_TABLE_FUNCTION_DECL(double, sinh_tbl_head);
+CLC_TABLE_FUNCTION_DECL(double, sinh_tbl_tail);
+CLC_TABLE_FUNCTION_DECL(double, cosh_tbl_head);
+CLC_TABLE_FUNCTION_DECL(double, cosh_tbl_tail);
 CLC_TABLE_FUNCTION_DECL(double, cbrt_inv_tbl);
 CLC_TABLE_FUNCTION_DECL(double, cbrt_dbl_tbl_head);
 CLC_TABLE_FUNCTION_DECL(double, cbrt_dbl_tbl_tail);

libclc/clc/lib/generic/SOURCES

@@ -31,6 +31,7 @@ math/clc_atanpi.cl
 math/clc_cbrt.cl
 math/clc_ceil.cl
 math/clc_copysign.cl
+math/clc_cosh.cl
 math/clc_cospi.cl
 math/clc_ep_log.cl
 math/clc_exp.cl
@@ -74,10 +75,12 @@ math/clc_rootn.cl
 math/clc_round.cl
 math/clc_rsqrt.cl
 math/clc_sincos_helpers.cl
+math/clc_sinh.cl
 math/clc_sinpi.cl
 math/clc_sqrt.cl
 math/clc_sw_fma.cl
 math/clc_tables.cl
+math/clc_tanh.cl
 math/clc_tanpi.cl
 math/clc_trunc.cl
 relational/clc_all.cl

libclc/clc/lib/generic/math/clc_cosh.cl

@@ -0,0 +1,24 @@
+//===----------------------------------------------------------------------===//
+//
+// 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/float/definitions.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_copysign.h>
+#include <clc/math/clc_exp.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_isinf.h>
+#include <clc/relational/clc_isnan.h>
+#include <clc/shared/clc_min.h>
+
+#define __CLC_BODY <clc_cosh.inc>
+#include <clc/math/gentype.inc>

libclc/clc/lib/generic/math/clc_cosh.inc

@@ -0,0 +1,199 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#if __CLC_FPSIZE == 32
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_cosh(__CLC_GENTYPE x) {
+  // After dealing with special cases the computation is split into regions as
+  // follows. abs(x) >= max_cosh_arg: cosh(x) = sign(x)*Inf abs(x) >=
+  // small_threshold: cosh(x) = sign(x)*exp(abs(x))/2 computed using the
+  // splitexp and scaleDouble functions as for exp_amd().
+  // abs(x) < small_threshold:
+  // compute p = exp(y) - 1 and then z = 0.5*(p+(p/(p+1.0)))
+  // cosh(x) is then z.
+
+  const __CLC_GENTYPE max_cosh_arg = 0x1.65a9fap+6f;
+  const __CLC_GENTYPE small_threshold = 0x1.0a2b24p+3f;
+
+  __CLC_UINTN ux = __CLC_AS_UINTN(x);
+  __CLC_GENTYPE y = __clc_fabs(x);
+  __CLC_UINTN aux = __CLC_AS_UINTN(y);
+
+  // Find the integer part y0 of y and the increment dy = y - y0. We then
+  // compute z = sinh(y) = sinh(y0)cosh(dy) + cosh(y0)sinh(dy) z = cosh(y) =
+  // cosh(y0)cosh(dy) + sinh(y0)sinh(dy) where sinh(y0) and cosh(y0) are
+  // tabulated above.
+
+  __CLC_INTN ind = __CLC_CONVERT_INTN(y);
+  ind = __CLC_CONVERT_UINTN(ind) > 36U ? 0 : ind;
+
+  __CLC_GENTYPE dy = y - __CLC_CONVERT_GENTYPE(ind);
+  __CLC_GENTYPE dy2 = dy * dy;
+
+  __CLC_GENTYPE sdy = __clc_mad(
+      dy2,
+      __clc_mad(
+          dy2,
+          __clc_mad(
+              dy2,
+              __clc_mad(
+                  dy2,
+                  __clc_mad(dy2,
+                            __clc_mad(dy2, 0.7746188980094184251527126e-12f,
+                                      0.160576793121939886190847e-9f),
+                            0.250521176994133472333666e-7f),
+                  0.275573191913636406057211e-5f),
+              0.198412698413242405162014e-3f),
+          0.833333333333329931873097e-2f),
+      0.166666666666666667013899e0f);
+  sdy = __clc_mad(sdy, dy * dy2, dy);
+
+  __CLC_GENTYPE cdy = __clc_mad(
+      dy2,
+      __clc_mad(
+          dy2,
+          __clc_mad(
+              dy2,
+              __clc_mad(
+                  dy2,
+                  __clc_mad(dy2,
+                            __clc_mad(dy2, 0.1163921388172173692062032e-10f,
+                                      0.208744349831471353536305e-8f),
+                            0.275573350756016588011357e-6f),
+                  0.248015872460622433115785e-4f),
+              0.138888888889814854814536e-2f),
+          0.416666666666660876512776e-1f),
+      0.500000000000000005911074e0f);
+  cdy = __clc_mad(cdy, dy2, 1.0f);
+
+  __CLC_GENTYPE sinhcoshh = USE_TABLE(sinhcosh_tbl_head, ind);
+  __CLC_GENTYPE sinhcosht = USE_TABLE(sinhcosh_tbl_tail, ind);
+  __CLC_GENTYPE z = __clc_mad(sinhcoshh, sdy, sinhcosht * cdy);
+
+  // When exp(-x) is insignificant compared to exp(x), return exp(x)/2
+  __CLC_GENTYPE t = __clc_exp(y - 0x1.62e500p-1f);
+  __CLC_GENTYPE zsmall = __clc_mad(0x1.a0210ep-18f, t, t);
+  z = y >= small_threshold ? zsmall : z;
+
+  // Corner cases
+  z = y >= max_cosh_arg ? __CLC_AS_GENTYPE((__CLC_UINTN)PINFBITPATT_SP32) : z;
+  z = aux > PINFBITPATT_SP32 ? __CLC_GENTYPE_NAN : z;
+  z = aux < 0x38800000 ? 1.0f : z;
+
+  return z;
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_cosh(__CLC_GENTYPE x) {
+  // After dealing with special cases the computation is split into
+  // regions as follows:
+  //
+  // abs(x) >= max_cosh_arg:
+  // cosh(x) = sign(x)*Inf
+  //
+  // abs(x) >= small_threshold:
+  // cosh(x) = sign(x)*exp(abs(x))/2 computed using the
+  // splitexp and scaleDouble functions as for exp_amd().
+  //
+  // abs(x) < small_threshold:
+  // compute p = exp(y) - 1 and then z = 0.5*(p+(p/(p+1.0)))
+  // cosh(x) is then sign(x)*z.
+
+  // This is ln(2^1025) = 0x408633ce8fb9f87e
+  const __CLC_GENTYPE max_cosh_arg = 7.10475860073943977113e+02;
+
+  // This is where exp(-x) is insignificant compared to exp(x) = ln(2^27)
+  const __CLC_GENTYPE small_threshold = 0x1.2b708872320e2p+4;
+
+  __CLC_GENTYPE y = __clc_fabs(x);
+
+  // In this range we find the integer part y0 of y
+  // and the increment dy = y - y0. We then compute
+  // z = cosh(y) = cosh(y0)cosh(dy) + sinh(y0)sinh(dy)
+  // where sinh(y0) and cosh(y0) are tabulated above.
+
+  __CLC_INTN ind = __clc_min(__CLC_CONVERT_INTN(y), 36);
+  __CLC_GENTYPE dy = y - __CLC_CONVERT_GENTYPE(ind);
+  __CLC_GENTYPE dy2 = dy * dy;
+
+  __CLC_GENTYPE sdy =
+      dy * dy2 *
+      __clc_fma(
+          dy2,
+          __clc_fma(
+              dy2,
+              __clc_fma(
+                  dy2,
+                  __clc_fma(
+                      dy2,
+                      __clc_fma(dy2,
+                                __clc_fma(dy2, 0.7746188980094184251527126e-12,
+                                          0.160576793121939886190847e-9),
+                                0.250521176994133472333666e-7),
+                      0.275573191913636406057211e-5),
+                  0.198412698413242405162014e-3),
+              0.833333333333329931873097e-2),
+          0.166666666666666667013899e0);
+
+  __CLC_GENTYPE cdy =
+      dy2 *
+      __clc_fma(
+          dy2,
+          __clc_fma(
+              dy2,
+              __clc_fma(
+                  dy2,
+                  __clc_fma(
+                      dy2,
+                      __clc_fma(dy2,
+                                __clc_fma(dy2, 0.1163921388172173692062032e-10,
+                                          0.208744349831471353536305e-8),
+                                0.275573350756016588011357e-6),
+                      0.248015872460622433115785e-4),
+                  0.138888888889814854814536e-2),
+              0.416666666666660876512776e-1),
+          0.500000000000000005911074e0);
+
+  // At this point sinh(dy) is approximated by dy + sdy,
+  // and cosh(dy) is approximated by 1 + cdy.
+  __CLC_GENTYPE cl = USE_TABLE(cosh_tbl_head, ind);
+  __CLC_GENTYPE ct = USE_TABLE(cosh_tbl_tail, ind);
+  __CLC_GENTYPE sl = USE_TABLE(sinh_tbl_head, ind);
+  __CLC_GENTYPE st = USE_TABLE(sinh_tbl_tail, ind);
+
+  __CLC_GENTYPE z =
+      __clc_fma(
+          sl, dy,
+          __clc_fma(sl, sdy,
+                    __clc_fma(cl, cdy,
+                              __clc_fma(st, dy, __clc_fma(st, sdy, ct * cdy)) +
+                                  ct))) +
+      cl;
+
+  // Other cases
+  z = y < 0x1.0p-28 ? 1.0 : z;
+
+  __CLC_GENTYPE t = __clc_exp(y - 0x1.62e42fefa3800p-1);
+  t = __clc_fma(t, -0x1.ef35793c76641p-45, t);
+  z = y >= small_threshold ? t : z;
+
+  z = y >= max_cosh_arg ? __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) : z;
+
+  z = __clc_isinf(x) || __clc_isnan(x) ? y : z;
+
+  return z;
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_cosh(__CLC_GENTYPE x) {
+  return __CLC_CONVERT_GENTYPE(__clc_cosh(__CLC_CONVERT_FLOATN(x)));
+}
+
+#endif

libclc/clc/lib/generic/math/clc_sinh.cl

@@ -0,0 +1,23 @@
+//===----------------------------------------------------------------------===//
+//
+// 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/internal/clc.h>
+#include <clc/math/clc_copysign.h>
+#include <clc/math/clc_exp.h>
+#include <clc/math/clc_fabs.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_isinf.h>
+#include <clc/relational/clc_isnan.h>
+#include <clc/shared/clc_min.h>
+
+#define __CLC_BODY <clc_sinh.inc>
+#include <clc/math/gentype.inc>