releaase 2.11 (#703)

This commit is contained in:
Aditya Atluri
2022-11-19 06:02:15 -08:00
committed by GitHub
parent 3c90f6aea6
commit c975e2ccbb
329 changed files with 47332 additions and 10607 deletions

View File

@@ -41,6 +41,8 @@
#include "cutlass/gemm/gemm.h"
#include "cutlass/conv/conv2d_problem_size.h"
#include "cutlass/conv/conv3d_problem_size.h"
#include "cutlass/gemm/threadblock/index_remat.h"
#include "cutlass/gemm/threadblock/threadblock_swizzle_streamk.h"
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -50,62 +52,6 @@ namespace threadblock {
/////////////////////////////////////////////////////////////////////////////////////////////////
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeThreadIdxX() {
return threadIdx.x;
}
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeThreadIdxY() {
return threadIdx.y;
}
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeThreadIdxZ() {
return threadIdx.z;
}
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockIdxX() {
return blockIdx.x;
}
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockIdxY() {
return blockIdx.y;
}
/// Helper to rematerialize block Idx. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockIdxZ() {
return blockIdx.z;
}
/// Helper to rematerialize block Dim. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockDimX() {
return blockDim.x;
}
/// Helper to rematerialize block Dim. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockDimY() {
return blockDim.y;
}
/// Helper to rematerialize block Dim. Reduces register liveness.
CUTLASS_DEVICE
int RematerializeBlockDimZ() {
return blockDim.z;
}
/////////////////////////////////////////////////////////////////////////////////////////////////
/// Threadblock swizzling function for GEMMs
template <int N = 1>
struct GemmIdentityThreadblockSwizzle {