[libclc] Move functions definition from header clc_sincos_piby4.inc into clc_sincos_helpers.cl (llvm#164028)

inline functions defined in clc_sincos_piby4.inc miss static specifier
and are deleted by EliminateAvailableExternallyPass when not inlined.

This PR fix the problem by removing inline and moving function
definition
into clc/lib/generic/math/clc_sincos_helpers.cl. It makes sense to put
all sin/cos helpers definitions in one file clc_sincos_helpers.cl.

Author: wenju-he

libclc/clc/include/clc/math/clc_sincos_helpers.inc

@@ -10,6 +10,11 @@ _CLC_DECL _CLC_OVERLOAD __CLC_FLOATN __clc_sinf_piby4(__CLC_FLOATN x,
                                                       __CLC_FLOATN y);
 _CLC_DECL _CLC_OVERLOAD __CLC_FLOATN __clc_cosf_piby4(__CLC_FLOATN x,
                                                       __CLC_FLOATN y);
+
+_CLC_DECL _CLC_OVERLOAD void __clc_sincos_piby4(__CLC_FLOATN x,
+                                                private __CLC_FLOATN *sinval,
+                                                private __CLC_FLOATN *cosval);
+
 _CLC_DECL _CLC_OVERLOAD __CLC_FLOATN __clc_tanf_piby4(__CLC_FLOATN x,
                                                       __CLC_INTN regn);
 

libclc/clc/include/clc/math/clc_sincos_helpers_fp64.inc

@@ -6,6 +6,15 @@
 //
 //===----------------------------------------------------------------------===//
 
+_CLC_DECL _CLC_OVERLOAD void __clc_sincos_piby4(__CLC_DOUBLEN x,
+                                                __CLC_DOUBLEN xx,
+                                                private __CLC_DOUBLEN *sinval,
+                                                private __CLC_DOUBLEN *cosval);
+
+_CLC_DECL _CLC_OVERLOAD void __clc_tan_piby4(__CLC_DOUBLEN x, __CLC_DOUBLEN xx,
+                                             private __CLC_DOUBLEN *leadval,
+                                             private __CLC_DOUBLEN *tailval);
+
 _CLC_DECL _CLC_OVERLOAD void
 __clc_remainder_piby2_medium(__CLC_DOUBLEN x, private __CLC_DOUBLEN *r,
                              private __CLC_DOUBLEN *rr,

libclc/clc/include/clc/math/clc_sincos_piby4.h

@@ -10,7 +10,6 @@
 #include <clc/float/definitions.h>
 #include <clc/math/clc_fabs.h>
 #include <clc/math/clc_sincos_helpers.h>
-#include <clc/math/clc_sincos_piby4.h>
 #include <clc/math/math.h>
 #include <clc/relational/clc_isinf.h>
 #include <clc/relational/clc_isnan.h>

libclc/clc/include/clc/math/clc_sincos_piby4.inc

@@ -11,7 +11,6 @@
 #include <clc/internal/clc.h>
 #include <clc/math/clc_fabs.h>
 #include <clc/math/clc_sincos_helpers.h>
-#include <clc/math/clc_sincos_piby4.h>
 #include <clc/math/math.h>
 
 #define __CLC_BODY <clc_cospi.inc>

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

@@ -11,7 +11,6 @@
 #include <clc/internal/clc.h>
 #include <clc/math/clc_fabs.h>
 #include <clc/math/clc_sincos_helpers.h>
-#include <clc/math/clc_sincos_piby4.h>
 #include <clc/math/clc_trunc.h>
 #include <clc/math/math.h>
 #include <clc/math/tables.h>

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

@@ -74,6 +74,43 @@ _CLC_DEF _CLC_OVERLOAD __CLC_FLOATN __clc_cosf_piby4(__CLC_FLOATN x,
   return ret;
 }
 
+// Evaluate single precisions sin and cos of value in interval [-pi/4, pi/4]
+_CLC_DEF _CLC_OVERLOAD void __clc_sincos_piby4(__CLC_FLOATN x,
+                                               private __CLC_FLOATN *sinval,
+                                               private __CLC_FLOATN *cosval) {
+  // Taylor series for sin(x) is x - x^3/3! + x^5/5! - x^7/7! ...
+  // = x * (1 - x^2/3! + x^4/5! - x^6/7! ...
+  // = x * f(w)
+  // where w = x*x and f(w) = (1 - w/3! + w^2/5! - w^3/7! ...
+  // We use a minimax approximation of (f(w) - 1) / w
+  // because this produces an expansion in even powers of x.
+
+  // Taylor series for cos(x) is 1 - x^2/2! + x^4/4! - x^6/6! ...
+  // = f(w)
+  // where w = x*x and f(w) = (1 - w/2! + w^2/4! - w^3/6! ...
+  // We use a minimax approximation of (f(w) - 1 + w/2) / (w*w)
+  // because this produces an expansion in even powers of x.
+
+  const __CLC_FLOATN sc1 = -0.166666666638608441788607926e0F;
+  const __CLC_FLOATN sc2 = 0.833333187633086262120839299e-2F;
+  const __CLC_FLOATN sc3 = -0.198400874359527693921333720e-3F;
+  const __CLC_FLOATN sc4 = 0.272500015145584081596826911e-5F;
+
+  const __CLC_FLOATN cc1 = 0.41666666664325175238031e-1F;
+  const __CLC_FLOATN cc2 = -0.13888887673175665567647e-2F;
+  const __CLC_FLOATN cc3 = 0.24800600878112441958053e-4F;
+  const __CLC_FLOATN cc4 = -0.27301013343179832472841e-6F;
+
+  __CLC_FLOATN x2 = x * x;
+
+  *sinval = __clc_mad(
+      x * x2, __clc_mad(x2, __clc_mad(x2, __clc_mad(x2, sc4, sc3), sc2), sc1),
+      x);
+  *cosval = __clc_mad(
+      x2 * x2, __clc_mad(x2, __clc_mad(x2, __clc_mad(x2, cc4, cc3), cc2), cc1),
+      __clc_mad(x2, -0.5f, 1.0f));
+}
+
 _CLC_DEF _CLC_OVERLOAD __CLC_FLOATN __clc_tanf_piby4(__CLC_FLOATN x,
                                                      __CLC_INTN regn) {
   // Core Remez [1,2] approximation to tan(x) on the interval [0,pi/4].