CUTLASS 3.2.1 (#1113)

* Updates for 3.2.1 release.

* Minor fix in gemm op profiler for raster order.

* Add scheduler mapping for raster order in the kernels.

include/cutlass/layout/matrix.h

@@ -37,6 +37,17 @@
     Layout functions must implement all members in the public interface of IdentityTensorLayout<>
     defined in cutlass/tensor_ref.h.
 */
+
+/*
+  Note:  CUTLASS 3x increases the host compiler requirements to C++17. However, certain
+         existing integrations of CUTLASS require C++11 host compilers.
+
+         Until this requirement can be lifted, certain headers with this annotation are required
+         to be remain consistent with C++11 syntax.
+
+         C++11 compatibility is enforced by this unit test: `cutlass_test_unit_core_cpp11`.
+*/
+
 #pragma once
 
 #include "cutlass/cutlass.h"
@@ -538,6 +549,7 @@ public:
   /// Inverse of layout function, mapping linear offset to logical coordinate
   CUTLASS_HOST_DEVICE
   MatrixCoord inverse(LongIndex offset) const {
+    CUTLASS_UNUSED(offset);
     return MatrixCoord(0, 0);
   }
 
@@ -796,6 +808,7 @@ public:
   /// Inverse of layout function, mapping linear offset to logical coordinate
   CUTLASS_HOST_DEVICE
   MatrixCoord inverse(LongIndex offset) const {
+    CUTLASS_UNUSED(offset);
     return MatrixCoord(0, 0);
   }
 
@@ -901,6 +914,7 @@ public:
   /// Inverse of layout function, mapping linear offset to logical coordinate
   CUTLASS_HOST_DEVICE
   MatrixCoord inverse(LongIndex offset) const {
+    CUTLASS_UNUSED(offset);
     return MatrixCoord(0, 0);
   }
 

include/cutlass/layout/pitch_linear.h

@@ -31,6 +31,17 @@
 /*! \file
     \brief Defines layout functions used by TensorRef and derived classes for pitch-linear memory.
 */
+
+/*
+  Note:  CUTLASS 3x increases the host compiler requirements to C++17. However, certain
+         existing integrations of CUTLASS require C++11 host compilers.
+
+         Until this requirement can be lifted, certain headers with this annotation are required
+         to be remain consistent with C++11 syntax.
+
+         C++11 compatibility is enforced by this unit test: `cutlass_test_unit_core_cpp11`.
+*/
+
 #pragma once
 
 #include "cutlass/cutlass.h"

include/cutlass/layout/vector.h

@@ -78,6 +78,7 @@ public:
   /// Helper returns a layout to a tightly packed tensor
   CUTLASS_HOST_DEVICE
   static PackedVectorLayout packed(TensorCoord const &size) {
+    CUTLASS_UNUSED(size);
     return PackedVectorLayout();
   }