CUTLASS 2.7 (#318)

CUTLASS 2.7

Mainloop fusion for GEMM: summation over A or B
Strided DGRAD (optimized iterators)
Half-precision GELU_taylor activation functions
Use these when accumulation and epilogue compute types are all cutlass::half_t
Tuning and bug fixes to fused GEMM + GEMM example
Support for smaller than 128b aligned Convolutions: see examples
Caching of results to accelerate Convolution unit tests
Can be enabled or disabled by running cmake .. -DCUTLASS_TEST_ENABLE_CACHED_RESULTS=OFF
Corrections and bug fixes reported by the CUTLASS community
Thank you for filing these issues!

authored-by: Haicheng Wu haichengw@nvidia.com, Manish Gupta manigupta@nvidia.com, Dustyn Blasig dblasig@nvidia.com, Andrew Kerr akerr@nvidia.com

include/cutlass/gemm/kernel/default_gemm_with_k_reduction.h

@@ -126,7 +126,7 @@ struct DefaultGemmWithKReduction {
           ThreadblockShape, typename Mma::Operator, kPartitionsK, EpilogueOutputOp,
           EpilogueOutputOp::kCount>::Epilogue;
 
-  /// Define the epilogue
+  /// Define the epilogue of the reduction vector
   using EpilogueGemmKReduction =
       typename cutlass::epilogue::threadblock::EpilogueGemmKReduction<
           ElementAccumulator, ElementC, ThreadblockShape, typename Mma::Operator, kReduceKForA>;

include/cutlass/gemm/kernel/gemm_with_k_reduction.h

@@ -582,6 +582,13 @@ public:
       __threadfence();
     }
 
+    // Execute the epilogue operator to update the destination tensor.
+    epilogue(
+      output_op, 
+      iterator_D, 
+      accumulators, 
+      iterator_C); 
+ 
     if ((kReduceKForA && threadblock_tile_offset.n() == 0)
      || (!kReduceKForA && threadblock_tile_offset.m() == 0)) {
 
@@ -610,14 +617,7 @@ public:
             && (threadblock_tile_offset.k() > 0));
       }
     }
-
-    // Execute the epilogue operator to update the destination tensor.
-    epilogue(
-      output_op, 
-      iterator_D, 
-      accumulators, 
-      iterator_C); 
-    
+   
     //
     // Release the semaphore
     //