[SYCL] Add external arbitrary floating point functions (#2114)

Add support of functions that operate on arbitrary precision floating
point numbers. This datatype and its corresponding operations can be
useful on targets that can take advantage of narrower representation
such as FPGAs.

Arbitrary precision floating point numbers are represented using
arbitrary precision integers (_ExtInt(N)). Additional parameters are:
* _E_: represents the number of exponent bits
* _M_: represents the number of mantissa bits
The total width of the type is `E+M+1` where the last bit is used to
represent the sign. The data layout is shown below:
`[ S (sign bit) ][ E (Exponent) ][ M (Mantissa) ]`

The width of the data `(E+M+1)` is encoded with the width of the
arbitrary precision integer variable. Mantissa (namely `M`) is encoded
within each of the signature of the operations. Exponent values
(namely `E`) are inferred from the width of corresponding variable.

Usage example:
  _ExtInt(Wa) a;
  _ExtInt(Wb) b;
  _ExtInt(Wout) res = __spirv_ArbitraryFloatAddINTEL<Wa, Wb, Wout>(a, Ma, b, Mb, Mout);

Signed-off-by: Mikhail Lychkov <[email protected]>

Author: mlychkov

sycl/include/CL/__spirv/spirv_ops.hpp

@@ -205,6 +205,14 @@ __SYCL_CONVERGENT__ extern SYCL_EXTERNAL void
 __spirv_SubgroupBlockWriteINTEL(__attribute__((opencl_global)) uint32_t *Ptr,
                                 dataT Data) noexcept;
 
+template <typename dataT>
+__SYCL_CONVERGENT__ extern SYCL_EXTERNAL dataT __spirv_SubgroupBlockReadINTEL(
+    const __attribute__((opencl_global)) uint64_t *Ptr) noexcept;
+
+template <typename dataT>
+__SYCL_CONVERGENT__ extern SYCL_EXTERNAL void
+__spirv_SubgroupBlockWriteINTEL(__attribute__((opencl_global)) uint64_t *Ptr,
+                                dataT Data) noexcept;
 template <int W, int rW>
 extern SYCL_EXTERNAL ap_int<rW>
 __spirv_FixedSqrtINTEL(ap_int<W> a, bool S, int32_t I, int32_t rI,
@@ -256,14 +264,232 @@ extern SYCL_EXTERNAL ap_int<rW>
 __spirv_FixedExpINTEL(ap_int<W> a, bool S, int32_t I, int32_t rI,
                       int32_t Quantization = 0, int32_t Overflow = 0) noexcept;
 
-template <typename dataT>
-__SYCL_CONVERGENT__ extern SYCL_EXTERNAL dataT __spirv_SubgroupBlockReadINTEL(
-    const __attribute__((opencl_global)) uint64_t *Ptr) noexcept;
-
-template <typename dataT>
-__SYCL_CONVERGENT__ extern SYCL_EXTERNAL void
-__spirv_SubgroupBlockWriteINTEL(__attribute__((opencl_global)) uint64_t *Ptr,
-                                dataT Data) noexcept;
+// In the following built-ins width of arbitrary precision integer type for
+// a floating point variable should be equal to sum of corresponding
+// exponent width E, mantissa width M and 1 for sign bit. I.e. WA = EA + MA + 1.
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCastINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCastFromIntINTEL(
+    ap_int<WA> A, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCastToIntINTEL(
+    ap_int<WA> A, int32_t MA, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatAddINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatSubINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatMulINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatDivINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+// Comparison built-ins don't use Subnormal Support, Rounding Mode and
+// Rounding Accuracy.
+template <int WA, int WB>
+extern SYCL_EXTERNAL bool
+__spirv_ArbitraryFloatGTINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                              int32_t MB) noexcept;
+
+template <int WA, int WB>
+extern SYCL_EXTERNAL bool
+__spirv_ArbitraryFloatGEINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                              int32_t MB) noexcept;
+
+template <int WA, int WB>
+extern SYCL_EXTERNAL bool
+__spirv_ArbitraryFloatLTINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                              int32_t MB) noexcept;
+
+template <int WA, int WB>
+extern SYCL_EXTERNAL bool
+__spirv_ArbitraryFloatLEINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                              int32_t MB) noexcept;
+
+template <int WA, int WB>
+extern SYCL_EXTERNAL bool
+__spirv_ArbitraryFloatEQINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                              int32_t MB) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatRecipINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatRSqrtINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCbrtINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatHypotINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatSqrtINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatLogINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatLog2INTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatLog10INTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatLog1pINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatExpINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatExp2INTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatExp10INTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatExpm1INTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatSinINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCosINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+// Result value contains both values of sine and cosine and so has the size of
+// 2 * Wout where Wout is equal to (1 + Eout + Mout).
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<2 * Wout> __spirv_ArbitraryFloatSinCosINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatSinPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatCosPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+// Result value contains both values of sine(A*pi) and cosine(A*pi) and so has
+// the size of 2 * Wout where Wout is equal to (1 + Eout + Mout).
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<2 * Wout> __spirv_ArbitraryFloatSinCosPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatASinINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatASinPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatACosINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatACosPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatATanINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatATanPiINTEL(
+    ap_int<WA> A, int32_t MA, int32_t Mout, int32_t EnableSubnormals = 0,
+    int32_t RoundingMode = 0, int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatATan2INTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatPowINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout> __spirv_ArbitraryFloatPowRINTEL(
+    ap_int<WA> A, int32_t MA, ap_int<WB> B, int32_t MB, int32_t Mout,
+    int32_t EnableSubnormals = 0, int32_t RoundingMode = 0,
+    int32_t RoundingAccuracy = 0) noexcept;
+
+// PowN built-in calculates `A^B` where `A` is arbitrary precision floating
+// point number and `B` is arbitrary precision integer, i.e. its width doesn't
+// depend on sum of exponent and mantissa.
+template <int WA, int WB, int Wout>
+extern SYCL_EXTERNAL ap_int<Wout>
+__spirv_ArbitraryFloatPowNINTEL(ap_int<WA> A, int32_t MA, ap_int<WB> B,
+                                int32_t Mout, int32_t EnableSubnormals = 0,
+                                int32_t RoundingMode = 0,
+                                int32_t RoundingAccuracy = 0) noexcept;
 
 template <typename dataT>
 extern SYCL_EXTERNAL int32_t __spirv_ReadPipe(RPipeTy<dataT> Pipe, dataT *Data,