[SYCL-PTX] Refactor of libclc sources in preparation of math builtins category implementation

The patch moves some include files from libs to include so they can be shared with libspirv side.
Refactor the ptx-nvidiacl side as well and move the following builtins in their own file to avoid symbol redefinition:
- acosh
- asinh
- atanh
- cosh
- erfc

Signed-off-by: Victor Lomuller <[email protected]>
Co-authored-by: David Wood <[email protected]>

Author: Naghasan

libclc/generic/include/clcmacro.h

@@ -1,3 +1,14 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_MACRO_H
+#define __CLC_MACRO_H
+
 #define _CLC_UNARY_VECTORIZE(DECLSPEC, RET_TYPE, FUNCTION, ARG1_TYPE)          \
   DECLSPEC RET_TYPE##2 FUNCTION(ARG1_TYPE##2 x) {                              \
     return (RET_TYPE##2)(FUNCTION(x.x), FUNCTION(x.y));                        \
@@ -177,3 +188,5 @@
 #define _CLC_DEFINE_UNARY_BUILTIN(RET_TYPE, FUNCTION, BUILTIN, ARG1_TYPE)      \
   _CLC_DEF _CLC_OVERLOAD RET_TYPE FUNCTION(ARG1_TYPE x) { return BUILTIN(x); } \
   _CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, RET_TYPE, FUNCTION, ARG1_TYPE)
+
+#endif // !__CLC_MACRO_H

libclc/ptx-nvidiacl/libspirv/math/binary_builtin.inc renamed to libclc/generic/include/math/binary_builtin.inc

@@ -6,11 +6,23 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "../../../generic/lib/clcmacro.h"
 #include "utils.h"
+#include <clcmacro.h>
+
+#ifndef __CLC_BUILTIN
+#define __CLC_BUILTIN __CLC_XCONCAT(__clc_, __CLC_FUNCTION)
+#endif
+
+#ifndef __CLC_BUILTIN_D
+#define __CLC_BUILTIN_D __CLC_BUILTIN
+#endif
 
 #ifndef __CLC_BUILTIN_F
-#define __CLC_BUILTIN_F __CLC_XCONCAT(__CLC_BUILTIN, f)
+#define __CLC_BUILTIN_F __CLC_BUILTIN
+#endif
+
+#ifndef __CLC_BUILTIN_H
+#define __CLC_BUILTIN_H __CLC_BUILTIN_F
 #endif
 
 _CLC_DEFINE_BINARY_BUILTIN(float, __CLC_FUNCTION, __CLC_BUILTIN_F, float, float)
@@ -21,7 +33,7 @@ _CLC_DEFINE_BINARY_BUILTIN(float, __CLC_FUNCTION, __CLC_BUILTIN_F, float, float)
 
 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
 
-_CLC_DEFINE_BINARY_BUILTIN(double, __CLC_FUNCTION, __CLC_BUILTIN, double,
+_CLC_DEFINE_BINARY_BUILTIN(double, __CLC_FUNCTION, __CLC_BUILTIN_D, double,
                            double)
 
 #endif
@@ -30,12 +42,8 @@ _CLC_DEFINE_BINARY_BUILTIN(double, __CLC_FUNCTION, __CLC_BUILTIN, double,
 
 #pragma OPENCL EXTENSION cl_khr_fp16 : enable
 
-_CLC_DEFINE_BINARY_BUILTIN(half, __CLC_FUNCTION, __CLC_BUILTIN, half, half)
+_CLC_DEFINE_BINARY_BUILTIN(half, __CLC_FUNCTION, __CLC_BUILTIN_H, half, half)
 
 #endif
 
 #endif
-
-#undef __CLC_BUILTIN
-#undef __CLC_BUILTIN_F
-#undef __CLC_FUNCTION

libclc/generic/include/math/math.h

@@ -0,0 +1,99 @@
+//===----------------------------------------------------------------------===//
+//
+// 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_H_
+#define __CLC_MATH_H_
+
+#include "as_type.h"
+#include "config.h"
+#include "func.h"
+
+#define SNAN 0x001
+#define QNAN 0x002
+#define NINF 0x004
+#define NNOR 0x008
+#define NSUB 0x010
+#define NZER 0x020
+#define PZER 0x040
+#define PSUB 0x080
+#define PNOR 0x100
+#define PINF 0x200
+
+#if (defined __AMDGCN__ || defined __R600__) && !defined __HAS_FMAF__
+#define HAVE_HW_FMA32() (0)
+#else
+#define HAVE_HW_FMA32() (1)
+#endif
+
+#define HAVE_BITALIGN() (0)
+#define HAVE_FAST_FMA32() (0)
+
+#define MATH_DIVIDE(X, Y) ((X) / (Y))
+#define MATH_RECIP(X) (1.0f / (X))
+#define MATH_SQRT(X) __spirv_ocl_sqrt(X)
+
+#define SIGNBIT_SP32 0x80000000
+#define EXSIGNBIT_SP32 0x7fffffff
+#define EXPBITS_SP32 0x7f800000
+#define MANTBITS_SP32 0x007fffff
+#define ONEEXPBITS_SP32 0x3f800000
+#define TWOEXPBITS_SP32 0x40000000
+#define HALFEXPBITS_SP32 0x3f000000
+#define IMPBIT_SP32 0x00800000
+#define QNANBITPATT_SP32 0x7fc00000
+#define INDEFBITPATT_SP32 0xffc00000
+#define PINFBITPATT_SP32 0x7f800000
+#define NINFBITPATT_SP32 0xff800000
+#define EXPBIAS_SP32 127
+#define EXPSHIFTBITS_SP32 23
+#define BIASEDEMIN_SP32 1
+#define EMIN_SP32 -126
+#define BIASEDEMAX_SP32 254
+#define EMAX_SP32 127
+#define LAMBDA_SP32 1.0e30
+#define MANTLENGTH_SP32 24
+#define BASEDIGITS_SP32 7
+
+_CLC_OVERLOAD _CLC_INLINE float __clc_flush_denormal_if_not_supported(float x) {
+  int ix = as_int(x);
+  if (!__clc_fp32_subnormals_supported() && ((ix & EXPBITS_SP32) == 0) &&
+      ((ix & MANTBITS_SP32) != 0)) {
+    ix &= SIGNBIT_SP32;
+    x = as_float(ix);
+  }
+  return x;
+}
+
+#ifdef cl_khr_fp64
+
+#define SIGNBIT_DP64 0x8000000000000000L
+#define EXSIGNBIT_DP64 0x7fffffffffffffffL
+#define EXPBITS_DP64 0x7ff0000000000000L
+#define MANTBITS_DP64 0x000fffffffffffffL
+#define ONEEXPBITS_DP64 0x3ff0000000000000L
+#define TWOEXPBITS_DP64 0x4000000000000000L
+#define HALFEXPBITS_DP64 0x3fe0000000000000L
+#define IMPBIT_DP64 0x0010000000000000L
+#define QNANBITPATT_DP64 0x7ff8000000000000L
+#define INDEFBITPATT_DP64 0xfff8000000000000L
+#define PINFBITPATT_DP64 0x7ff0000000000000L
+#define NINFBITPATT_DP64 0xfff0000000000000L
+#define EXPBIAS_DP64 1023
+#define EXPSHIFTBITS_DP64 52
+#define BIASEDEMIN_DP64 1
+#define EMIN_DP64 -1022
+#define BIASEDEMAX_DP64 2046 /* 0x7fe */
+#define EMAX_DP64 1023       /* 0x3ff */
+#define LAMBDA_DP64 1.0e300
+#define MANTLENGTH_DP64 53
+#define BASEDIGITS_DP64 15
+
+#endif // cl_khr_fp64
+
+#define ALIGNED(x) __attribute__((aligned(x)))
+#endif // __CLC_MATH_H_

libclc/ptx-nvidiacl/libspirv/math/unary_builtin.inc renamed to libclc/generic/include/math/unary_builtin.inc

@@ -6,11 +6,23 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "../../../generic/lib/clcmacro.h"
 #include "utils.h"
+#include <clcmacro.h>
+
+#ifndef __CLC_BUILTIN
+#define __CLC_BUILTIN __CLC_XCONCAT(__clc_, __CLC_FUNCTION)
+#endif
+
+#ifndef __CLC_BUILTIN_D
+#define __CLC_BUILTIN_D __CLC_BUILTIN
+#endif
 
 #ifndef __CLC_BUILTIN_F
-#define __CLC_BUILTIN_F __CLC_XCONCAT(__CLC_BUILTIN, f)
+#define __CLC_BUILTIN_F __CLC_BUILTIN
+#endif
+
+#ifndef __CLC_BUILTIN_H
+#define __CLC_BUILTIN_H __CLC_BUILTIN_F
 #endif
 
 _CLC_DEFINE_UNARY_BUILTIN(float, __CLC_FUNCTION, __CLC_BUILTIN_F, float)
@@ -21,20 +33,16 @@ _CLC_DEFINE_UNARY_BUILTIN(float, __CLC_FUNCTION, __CLC_BUILTIN_F, float)
 
 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
 
-_CLC_DEFINE_UNARY_BUILTIN(double, __CLC_FUNCTION, __CLC_BUILTIN, double)
+_CLC_DEFINE_UNARY_BUILTIN(double, __CLC_FUNCTION, __CLC_BUILTIN_D, double)
 
 #endif
 
 #ifdef cl_khr_fp16
 
 #pragma OPENCL EXTENSION cl_khr_fp16 : enable
 
-_CLC_DEFINE_UNARY_BUILTIN(half, __CLC_FUNCTION, __CLC_BUILTIN, half)
-
-#endif
+_CLC_DEFINE_UNARY_BUILTIN(half, __CLC_FUNCTION, __CLC_BUILTIN_H, half)
 
 #endif
 
-#undef __CLC_BUILTIN
-#undef __CLC_BUILTIN_F
-#undef __CLC_FUNCTION
+#endif // !__FLOAT_ONLY

libclc/generic/lib/clcmacro.h

@@ -0,0 +1 @@
+../include/clcmacro.h

libclc/generic/lib/clcmacro.h

@@ -23,7 +23,7 @@
 #include <clc/clc.h>
 #include <spirv/spirv.h>
 
-#include "../clcmacro.h"
+#include <clcmacro.h>
 
 _CLC_OVERLOAD _CLC_DEF float degrees(float radians) {
   return __spirv_ocl_degrees(radians);