CUTLASS 2.1 (#83)

CUTLASS 2.1 contributes:
- BLAS-style host-side API added to CUTLASS Library
- Planar Complex GEMM kernels targeting Volta and Turing Tensor Cores
- Minor enhancements and bug fixes

include/cutlass/complex.h

@@ -25,7 +25,11 @@
 #pragma once
 
 #include <cuComplex.h>
+#if defined(__CUDACC_RTC__)
+#include <cuda/std/cstdint>
+#else
 #include <cstdint>
+#endif
 
 #include "cutlass/cutlass.h"
 #include "cutlass/half.h"
@@ -351,6 +355,16 @@ CUTLASS_HOST_DEVICE R norm_accumulate(complex<T> const &z, R const &accumulator)
     static_cast<R>(imag(z)) * static_cast<R>(imag(z));
 }
 
+/// Returns the complex conjugate
+CUTLASS_HOST_DEVICE float conj(float const &z) {
+  return z;
+}
+
+/// Returns the complex conjugate
+CUTLASS_HOST_DEVICE double conj(double const &z) {
+  return z;
+}
+
 /// Returns the complex conjugate
 template <typename T>
 CUTLASS_HOST_DEVICE complex<T> conj(complex<T> const &z) {
@@ -414,6 +428,10 @@ CUTLASS_HOST_DEVICE complex<T> sin(complex<T> const &z) {
 template <typename T>
 struct RealType< complex<T> > {
   using Type = T;
+
+  static complex<T> from_real(double x) {
+    return complex<T>(static_cast<T>(x));
+  }
 };
 
 /////////////////////////////////////////////////////////////////////////////////////////////////
@@ -438,5 +456,18 @@ cutlass::complex<double> from_real<cutlass::complex<double> >(double r) {
 
 //////////////////////////////////////////////////////////////////////////////////////////////////
 
+template <typename T>
+struct is_complex {
+  static bool const value = false;
+};
+
+template <typename T>
+struct is_complex<complex<T>> {
+  static bool const value = true;
+};
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+
 }  // namespace cutlass
 
+//////////////////////////////////////////////////////////////////////////////////////////////////