[libclc] Move exp, exp2 and expm1 to the CLC library

libclc/generic/lib/math/exp_helper.h → libclc/clc/include/clc/math/clc_exp.h

@@ -6,9 +6,15 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifdef cl_khr_fp64
+#ifndef __CLC_MATH_CLC_EXP_H__
+#define __CLC_MATH_CLC_EXP_H__
 
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable
-_CLC_DECL double __clc_exp_helper(double x, double x_min, double x_max, double r, int n);
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_exp
 
-#endif
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_EXP_H__

libclc/clc/include/clc/math/clc_exp2.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_EXP2_H__
+#define __CLC_MATH_CLC_EXP2_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_exp2
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_EXP2_H__

libclc/clc/include/clc/math/clc_exp_helper.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_EXP_HELPER
+#define __CLC_MATH_CLC_EXP_HELPER
+
+#define __DOUBLE_ONLY
+#define __CLC_BODY <clc/math/clc_exp_helper.inc>
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __DOUBLE_ONLY
+
+#endif // __CLC_MATH_CLC_EXP_HELPER

libclc/clc/include/clc/math/clc_exp_helper.inc

@@ -0,0 +1,13 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_DECL _CLC_OVERLOAD __CLC_GENTYPE __clc_exp_helper(__CLC_GENTYPE x,
+                                                       __CLC_GENTYPE x_min,
+                                                       __CLC_GENTYPE x_max,
+                                                       __CLC_GENTYPE r,
+                                                       __CLC_INTN n);

libclc/clc/include/clc/math/clc_expm1.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_EXPM1_H__
+#define __CLC_MATH_CLC_EXPM1_H__
+
+#define __CLC_BODY <clc/math/unary_decl.inc>
+#define __CLC_FUNCTION __clc_expm1
+
+#include <clc/math/gentype.inc>
+
+#undef __CLC_BODY
+#undef __CLC_FUNCTION
+
+#endif // __CLC_MATH_CLC_EXPM1_H__

libclc/clc/lib/generic/SOURCES

@@ -32,7 +32,11 @@ math/clc_ceil.cl
 math/clc_copysign.cl
 math/clc_cospi.cl
 math/clc_ep_log.cl
+math/clc_exp.cl
 math/clc_exp10.cl
+math/clc_exp2.cl
+math/clc_expm1.cl
+math/clc_exp_helper.cl
 math/clc_fabs.cl
 math/clc_fma.cl
 math/clc_fmod.cl

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

@@ -0,0 +1,19 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_exp_helper.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/math.h>
+#include <clc/relational/clc_isnan.h>
+
+#define __CLC_BODY <clc_exp.inc>
+#include <clc/math/gentype.inc>

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

@@ -0,0 +1,76 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_exp(__CLC_GENTYPE x) {
+  // Reduce x
+  const __CLC_GENTYPE ln2HI = 0x1.62e300p-1f;
+  const __CLC_GENTYPE ln2LO = 0x1.2fefa2p-17f;
+  const __CLC_GENTYPE invln2 = 0x1.715476p+0f;
+
+  __CLC_GENTYPE fhalF = x < 0.0f ? -0.5f : 0.5f;
+  __CLC_INTN p = __CLC_CONVERT_INTN(__clc_mad(x, invln2, fhalF));
+  __CLC_GENTYPE fp = __CLC_CONVERT_GENTYPE(p);
+  __CLC_GENTYPE hi = __clc_mad(fp, -ln2HI, x); // t*ln2HI is exact here
+  __CLC_GENTYPE lo = -fp * ln2LO;
+
+  // Evaluate poly
+  __CLC_GENTYPE t = hi + lo;
+  __CLC_GENTYPE tt = t * t;
+  __CLC_GENTYPE v = __clc_mad(
+      tt,
+      -__clc_mad(
+          tt,
+          __clc_mad(tt,
+                    __clc_mad(tt,
+                              __clc_mad(tt, 0x1.637698p-25f, -0x1.bbd41cp-20f),
+                              0x1.1566aap-14f),
+                    -0x1.6c16c2p-9f),
+          0x1.555556p-3f),
+      t);
+
+  __CLC_GENTYPE y = 1.0f - (((-lo) - MATH_DIVIDE(t * v, 2.0f - v)) - hi);
+
+  // Scale by 2^p
+  __CLC_GENTYPE r = __CLC_AS_GENTYPE(__CLC_AS_INTN(y) + (p << 23));
+
+  // ln(largest_normal) = 88.72283905206835305366
+  const __CLC_GENTYPE ulim = 0x1.62e430p+6f;
+  // ln(smallest_normal) = -87.33654475055310898657
+  const __CLC_GENTYPE llim = -0x1.5d589ep+6f;
+
+  r = x < llim ? 0.0f : r;
+  r = x < ulim ? r : __CLC_AS_GENTYPE((__CLC_UINTN)0x7f800000);
+  return __clc_isnan(x) ? x : r;
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp(__CLC_GENTYPE x) {
+
+  const __CLC_GENTYPE X_MIN = -0x1.74910d52d3051p+9;           // -1075*ln(2)
+  const __CLC_GENTYPE X_MAX = 0x1.62e42fefa39efp+9;            // 1024*ln(2)
+  const __CLC_GENTYPE R_64_BY_LOG2 = 0x1.71547652b82fep+6;     // 64/ln(2)
+  const __CLC_GENTYPE R_LOG2_BY_64_LD = 0x1.62e42fefa0000p-7;  // head ln(2)/64
+  const __CLC_GENTYPE R_LOG2_BY_64_TL = 0x1.cf79abc9e3b39p-46; // tail ln(2)/64
+
+  __CLC_INTN n = __CLC_CONVERT_INTN(x * R_64_BY_LOG2);
+  __CLC_GENTYPE r =
+      __clc_fma(-R_LOG2_BY_64_TL, __CLC_CONVERT_GENTYPE(n),
+                __clc_fma(-R_LOG2_BY_64_LD, __CLC_CONVERT_GENTYPE(n), x));
+  return __clc_exp_helper(x, X_MIN, X_MAX, r, n);
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp(__CLC_GENTYPE x) {
+  return __CLC_CONVERT_GENTYPE(__clc_exp(__CLC_CONVERT_FLOATN(x)));
+}
+
+#endif

libclc/clc/lib/generic/math/clc_exp2.cl

@@ -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
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/float/definitions.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_exp_helper.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_mad.h>
+#include <clc/math/clc_rint.h>
+#include <clc/math/math.h>
+#include <clc/relational/clc_isnan.h>
+
+#define __CLC_BODY <clc_exp2.inc>
+#include <clc/math/gentype.inc>

libclc/clc/lib/generic/math/clc_exp2.inc

@@ -0,0 +1,68 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_exp2(__CLC_GENTYPE x) {
+  // Reduce x
+  const __CLC_GENTYPE ln2HI = 0x1.62e300p-1f;
+  const __CLC_GENTYPE ln2LO = 0x1.2fefa2p-17f;
+
+  __CLC_GENTYPE t = __clc_rint(x);
+  __CLC_INTN p = __CLC_CONVERT_INTN(t);
+  __CLC_GENTYPE tt = x - t;
+  __CLC_GENTYPE hi = tt * ln2HI;
+  __CLC_GENTYPE lo = tt * ln2LO;
+
+  // Evaluate poly
+  t = hi + lo;
+  tt = t * t;
+  __CLC_GENTYPE v = __clc_mad(
+      tt,
+      -__clc_mad(
+          tt,
+          __clc_mad(tt,
+                    __clc_mad(tt,
+                              __clc_mad(tt, 0x1.637698p-25f, -0x1.bbd41cp-20f),
+                              0x1.1566aap-14f),
+                    -0x1.6c16c2p-9f),
+          0x1.555556p-3f),
+      t);
+
+  __CLC_GENTYPE y = 1.0f - (((-lo) - MATH_DIVIDE(t * v, 2.0f - v)) - hi);
+
+  // Scale by 2^p
+  __CLC_GENTYPE r = __CLC_AS_FLOATN(__CLC_AS_INTN(y) + (p << 23));
+
+  const __CLC_GENTYPE ulim = 128.0f;
+  const __CLC_GENTYPE llim = -126.0f;
+
+  r = x < llim ? 0.0f : r;
+  r = x < ulim ? r : __CLC_AS_FLOATN((__CLC_UINTN)0x7f800000);
+  return __clc_isnan(x) ? x : r;
+}
+
+#elif __CLC_FPSIZE == 64
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp2(__CLC_GENTYPE x) {
+  const __CLC_GENTYPE R_LN2 = 0x1.62e42fefa39efp-1; // ln(2)
+  const __CLC_GENTYPE R_1_BY_64 = 1.0 / 64.0;
+
+  __CLC_INTN n = __CLC_CONVERT_INTN(x * 64.0);
+  __CLC_GENTYPE r = R_LN2 * __clc_fma(-R_1_BY_64, __CLC_CONVERT_GENTYPE(n), x);
+
+  return __clc_exp_helper(x, -1074.0, 1024.0, r, n);
+}
+
+#elif __CLC_FPSIZE == 16
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_exp2(__CLC_GENTYPE x) {
+  return __CLC_CONVERT_GENTYPE(__clc_exp2(__CLC_CONVERT_FLOATN(x)));
+}
+
+#endif

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

@@ -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
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_fma.h>
+#include <clc/math/clc_ldexp.h>
+#include <clc/math/math.h>
+#include <clc/math/tables.h>
+#include <clc/relational/clc_isnan.h>
+
+#define __DOUBLE_ONLY
+#define __CLC_BODY <clc_exp_helper.inc>
+
+#include <clc/math/gentype.inc>

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

@@ -0,0 +1,54 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_DEF _CLC_OVERLOAD __CLC_GENTYPE __clc_exp_helper(__CLC_GENTYPE x,
+                                                      __CLC_GENTYPE x_min,
+                                                      __CLC_GENTYPE x_max,
+                                                      __CLC_GENTYPE r,
+                                                      __CLC_INTN n) {
+
+  __CLC_INTN j = n & 0x3f;
+  __CLC_INTN m = n >> 6;
+
+  // 6 term tail of Taylor expansion of e^r
+  __CLC_GENTYPE z2 =
+      r * __clc_fma(
+              r,
+              __clc_fma(r,
+                        __clc_fma(r,
+                                  __clc_fma(r,
+                                            __clc_fma(r, 0x1.6c16c16c16c17p-10,
+                                                      0x1.1111111111111p-7),
+                                            0x1.5555555555555p-5),
+                                  0x1.5555555555555p-3),
+                        0x1.0000000000000p-1),
+              1.0);
+
+  __CLC_GENTYPE tv0 = USE_TABLE(two_to_jby64_ep_tbl_head, j);
+  __CLC_GENTYPE tv1 = USE_TABLE(two_to_jby64_ep_tbl_tail, j);
+  z2 = __clc_fma(tv0 + tv1, z2, tv1) + tv0;
+
+  __CLC_INTN small_value =
+      (m < -1022) || ((m == -1022) && __CLC_CONVERT_INTN(z2 < 1.0));
+
+  __CLC_INTN n1 = m >> 2;
+  __CLC_INTN n2 = m - n1;
+  __CLC_GENTYPE z3 =
+      z2 * __CLC_AS_GENTYPE((__CLC_CONVERT_LONGN(n1) + 1023) << 52);
+  z3 *= __CLC_AS_GENTYPE((__CLC_CONVERT_LONGN(n2) + 1023) << 52);
+
+  z2 = __clc_ldexp(z2, m);
+  z2 = __CLC_CONVERT_LONGN(small_value) ? z3 : z2;
+
+  z2 = __clc_isnan(x) ? x : z2;
+
+  z2 = x > x_max ? __CLC_AS_GENTYPE((__CLC_ULONGN)PINFBITPATT_DP64) : z2;
+  z2 = x < x_min ? 0.0 : z2;
+
+  return z2;
+}

libclc/clc/lib/generic/math/clc_expm1.cl

@@ -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
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/clc_convert.h>
+#include <clc/float/definitions.h>
+#include <clc/internal/clc.h>
+#include <clc/math/clc_exp_helper.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_isnan.h>
+
+#define __CLC_BODY <clc_expm1.inc>
+#include <clc/math/gentype.inc>