[libc][math] Refactor cosf16 implementation to header-only in src/__support/math folder.

libc/shared/math.h

@@ -34,6 +34,7 @@
 #include "math/cbrtf.h"
 #include "math/cos.h"
 #include "math/cosf.h"
+#include "math/cosf16.h"
 #include "math/erff.h"
 #include "math/exp.h"
 #include "math/exp10.h"

libc/shared/math/cosf16.h

@@ -0,0 +1,28 @@
+//===-- Shared cosf16 function ----------------------------------*- C++ -*-===//
+//
+// 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 LLVM_LIBC_SHARED_MATH_COSF16_H
+#define LLVM_LIBC_SHARED_MATH_COSF16_H
+
+#include "shared/libc_common.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "src/__support/math/cosf16.h"
+
+namespace LIBC_NAMESPACE_DECL {
+namespace shared {
+
+using math::cosf16;
+
+} // namespace shared
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LIBC_TYPES_HAS_FLOAT16
+
+#endif // LLVM_LIBC_SHARED_MATH_COSF16_H

libc/src/__support/math/CMakeLists.txt

@@ -390,6 +390,23 @@ add_header_library(
     libc.src.__support.macros.optimization
 )
 
+add_header_library(
+  cosf16
+  HDRS
+    cosf16.h
+  DEPENDS
+    .sincosf16_utils
+    libc.hdr.errno_macros
+    libc.hdr.fenv_macros
+    libc.src.__support.FPUtil.cast
+    libc.src.__support.FPUtil.fenv_impl
+    libc.src.__support.FPUtil.fp_bits
+    libc.src.__support.FPUtil.except_value_utils
+    libc.src.__support.FPUtil.multiply_add
+    libc.src.__support.macros.optimization
+    libc.src.__support.macros.properties.types
+)
+
 add_header_library(
   erff
   HDRS
@@ -699,3 +716,13 @@ add_header_library(
     libc.src.__support.FPUtil.polyeval
     libc.src.__support.common
 )
+
+add_header_library(
+  sincosf16_utils
+  HDRS
+    sincosf16_utils.h
+  DEPENDS
+    libc.src.__support.FPUtil.polyeval
+    libc.src.__support.FPUtil.nearest_integer
+    libc.src.__support.common
+)

libc/src/__support/math/cosf16.h

@@ -0,0 +1,106 @@
+//===-- Implementation header for cosf16 ------------------------*- C++ -*-===//
+//
+// 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 LLVM_LIBC_SRC___SUPPORT_MATH_COSF16_H
+#define LLVM_LIBC_SRC___SUPPORT_MATH_COSF16_H
+
+#include "include/llvm-libc-macros/float16-macros.h"
+
+#ifdef LIBC_TYPES_HAS_FLOAT16
+
+#include "sincosf16_utils.h"
+#include "src/__support/FPUtil/FEnvImpl.h"
+#include "src/__support/FPUtil/FPBits.h"
+#include "src/__support/FPUtil/cast.h"
+#include "src/__support/FPUtil/except_value_utils.h"
+#include "src/__support/FPUtil/multiply_add.h"
+#include "src/__support/macros/optimization.h"
+
+namespace LIBC_NAMESPACE_DECL {
+
+namespace math {
+
+LIBC_INLINE static constexpr float16 cosf16(float16 x) {
+#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+  constexpr size_t N_EXCEPTS = 4;
+
+  constexpr fputil::ExceptValues<float16, N_EXCEPTS> COSF16_EXCEPTS{{
+      // (input, RZ output, RU offset, RD offset, RN offset)
+      {0x2b7c, 0x3bfc, 1, 0, 1},
+      {0x4ac1, 0x38b5, 1, 0, 0},
+      {0x5c49, 0xb8c6, 0, 1, 0},
+      {0x7acc, 0xa474, 0, 1, 0},
+  }};
+#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+
+  using namespace sincosf16_internal;
+  using FPBits = fputil::FPBits<float16>;
+  FPBits xbits(x);
+
+  uint16_t x_u = xbits.uintval();
+  uint16_t x_abs = x_u & 0x7fff;
+  float xf = x;
+
+  // Range reduction:
+  // For |x| > pi/32, we perform range reduction as follows:
+  // Find k and y such that:
+  //   x = (k + y) * pi/32
+  //   k is an integer, |y| < 0.5
+  //
+  // This is done by performing:
+  //   k = round(x * 32/pi)
+  //   y = x * 32/pi - k
+  //
+  // Once k and y are computed, we then deduce the answer by the cosine of sum
+  // formula:
+  //   cos(x) = cos((k + y) * pi/32)
+  //          = cos(k * pi/32) * cos(y * pi/32) -
+  //            sin(k * pi/32) * sin(y * pi/32)
+
+#ifndef LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+  // Handle exceptional values
+  if (auto r = COSF16_EXCEPTS.lookup(x_abs); LIBC_UNLIKELY(r.has_value()))
+    return r.value();
+#endif // !LIBC_MATH_HAS_SKIP_ACCURATE_PASS
+
+  // cos(+/-0) = 1
+  if (LIBC_UNLIKELY(x_abs == 0U))
+    return fputil::cast<float16>(1.0f);
+
+  // cos(+/-inf) = NaN, and cos(NaN) = NaN
+  if (xbits.is_inf_or_nan()) {
+    if (xbits.is_signaling_nan()) {
+      fputil::raise_except_if_required(FE_INVALID);
+      return FPBits::quiet_nan().get_val();
+    }
+
+    if (xbits.is_inf()) {
+      fputil::set_errno_if_required(EDOM);
+      fputil::raise_except_if_required(FE_INVALID);
+    }
+
+    return x + FPBits::quiet_nan().get_val();
+  }
+
+  float sin_k = 0.0f, cos_k = 0.0f, sin_y = 0.0f, cosm1_y = 0.0f;
+  sincosf16_eval(xf, sin_k, cos_k, sin_y, cosm1_y);
+  // Since, cosm1_y = cos_y - 1, therefore:
+  //   cos(x) = cos_k * cos_y - sin_k * sin_y
+  //          = cos_k * (cos_y - 1 + 1) - sin_k * sin_y
+  //          = cos_k * cosm1_y - sin_k * sin_y + cos_k
+  return fputil::cast<float16>(fputil::multiply_add(
+      cos_k, cosm1_y, fputil::multiply_add(-sin_k, sin_y, cos_k)));
+}
+
+} // namespace math
+
+} // namespace LIBC_NAMESPACE_DECL
+
+#endif // LIBC_TYPES_HAS_FLOAT16
+
+#endif // LLVM_LIBC_SRC___SUPPORT_MATH_COSF16_H

libc/src/math/generic/sincosf16_utils.h → libc/src/__support/math/sincosf16_utils.h

@@ -16,6 +16,8 @@
 
 namespace LIBC_NAMESPACE_DECL {
 
+namespace sincosf16_internal {
+
 // Lookup table for sin(k * pi / 32) with k = 0, ..., 63.
 // Table is generated with Sollya as follows:
 // > display = hexadecimmal;
@@ -66,7 +68,7 @@ LIBC_INLINE int32_t range_reduction_sincosf16(float x, float &y) {
   return static_cast<int32_t>(kd);
 }
 
-static LIBC_INLINE void sincosf16_poly_eval(int32_t k, float y, float &sin_k,
+LIBC_INLINE static void sincosf16_poly_eval(int32_t k, float y, float &sin_k,
                                             float &cos_k, float &sin_y,
                                             float &cosm1_y) {
 
@@ -107,6 +109,8 @@ LIBC_INLINE void sincospif16_eval(float xf, float &sin_k, float &cos_k,
   sincosf16_poly_eval(k, y, sin_k, cos_k, sin_y, cosm1_y);
 }
 
+} // namespace sincosf16_internal
+
 } // namespace LIBC_NAMESPACE_DECL
 
 #endif // LLVM_LIBC_SRC_MATH_GENERIC_SINCOSF16_UTILS_H

libc/src/math/generic/CMakeLists.txt

@@ -278,16 +278,6 @@ add_entrypoint_object(
     libc.src.__support.FPUtil.generic.add_sub
 )
 
-add_header_library(
-  sincosf16_utils
-  HDRS
-    sincosf16_utils.h
-  DEPENDS
-    libc.src.__support.FPUtil.polyeval
-    libc.src.__support.FPUtil.nearest_integer
-    libc.src.__support.common
-)
-
 add_entrypoint_object(
   cos
   SRCS
@@ -315,16 +305,7 @@ add_entrypoint_object(
   HDRS
     ../cosf16.h
   DEPENDS
-    .sincosf16_utils
-    libc.hdr.errno_macros
-    libc.hdr.fenv_macros
-    libc.src.__support.FPUtil.cast
-    libc.src.__support.FPUtil.fenv_impl
-    libc.src.__support.FPUtil.fp_bits
-    libc.src.__support.FPUtil.except_value_utils
-    libc.src.__support.FPUtil.multiply_add
-    libc.src.__support.macros.optimization
-    libc.src.__support.macros.properties.types
+    libc.src.__support.math.cosf16
 )
 
 add_entrypoint_object(
@@ -349,14 +330,14 @@ add_entrypoint_object(
   HDRS
     ../cospif16.h
   DEPENDS
-    .sincosf16_utils
     libc.hdr.errno_macros
     libc.hdr.fenv_macros
     libc.src.__support.FPUtil.cast
     libc.src.__support.FPUtil.fenv_impl
     libc.src.__support.FPUtil.fp_bits
     libc.src.__support.FPUtil.multiply_add
     libc.src.__support.macros.optimization
+    libc.src.__support.math.sincosf16_utils
 )
 
 add_entrypoint_object(
@@ -405,7 +386,6 @@ add_entrypoint_object(
   HDRS
     ../sinf16.h
   DEPENDS
-    .sincosf16_utils
     libc.hdr.errno_macros
     libc.hdr.fenv_macros
     libc.src.__support.FPUtil.cast
@@ -415,6 +395,7 @@ add_entrypoint_object(
     libc.src.__support.FPUtil.multiply_add
     libc.src.__support.macros.optimization
     libc.src.__support.macros.properties.types
+    libc.src.__support.math.sincosf16_utils
   COMPILE_OPTIONS
     ${libc_opt_high_flag}
 )
@@ -482,14 +463,14 @@ add_entrypoint_object(
   HDRS
     ../sinpif16.h
   DEPENDS
-    .sincosf16_utils
     libc.hdr.errno_macros
     libc.hdr.fenv_macros
     libc.src.__support.FPUtil.cast
     libc.src.__support.FPUtil.fenv_impl
     libc.src.__support.FPUtil.fp_bits
     libc.src.__support.FPUtil.multiply_add
     libc.src.__support.macros.optimization
+    libc.src.__support.math.sincosf16_utils
 )
 
 add_entrypoint_object(
@@ -538,7 +519,6 @@ add_entrypoint_object(
   HDRS
     ../tanf16.h
   DEPENDS
-    .sincosf16_utils
     libc.hdr.errno_macros
     libc.hdr.fenv_macros
     libc.src.__support.FPUtil.cast
@@ -548,6 +528,7 @@ add_entrypoint_object(
     libc.src.__support.FPUtil.multiply_add
     libc.src.__support.macros.optimization
     libc.src.__support.macros.properties.types
+    libc.src.__support.math.sincosf16_utils
 )
 
 add_entrypoint_object(
@@ -572,7 +553,6 @@ add_entrypoint_object(
   HDRS
     ../tanpif16.h
   DEPENDS
-    .sincosf16_utils
     libc.hdr.errno_macros
     libc.hdr.fenv_macros
     libc.src.__support.FPUtil.cast
@@ -581,6 +561,7 @@ add_entrypoint_object(
     libc.src.__support.FPUtil.except_value_utils
     libc.src.__support.FPUtil.multiply_add
     libc.src.__support.macros.optimization
+    libc.src.__support.math.sincosf16_utils
 )
 
 add_entrypoint_object(