leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Fix typos 2 (#842)

Browse Source 
Co-authored-by: Haicheng Wu <57973641+hwu36@users.noreply.github.com>
This commit is contained in:
Alexander Pivovarov
2023-03-09 20:22:56 -08:00
committed by GitHub
parent c4f6b8c6bc
commit 7e370c9637
161 changed files with 310 additions and 309 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
include/cutlass/conv/threadblock/conv2d_dgrad_output_gradient_tile_access_iterator_optimized.h
View File

@@ -230,7 +230,7 @@ public:
offset_p[s] = (mapped_h + problem_size_.pad_h - filter_r) / problem_size_.stride_h;
offset_q[s] = (mapped_w + problem_size_.pad_w - filter_s) / problem_size_.stride_w;
// Intialize pointers for gemm_k=0
// Initialize pointers for gemm_k=0
TensorCoord coord{offset_n[s], offset_p[s], offset_q[s], filter_k_};
pointer_[s] += params_.layout(coord) * sizeof_bits<Element>::value / 8;
@@ -341,7 +341,7 @@ public:
next_idx = 1;
// Restore bytes in q coordinate (Mma in filter s dimenstion)
// Restore bytes in q coordinate (Mma in filter s dimension)
reset_bytes = reset_bytes_s_;
} else {
@@ -351,7 +351,7 @@ public:
next_idx = 2;
// Restore bytes in p and q coordinate (Mma in filter s and r dimenstion)
// Restore bytes in p and q coordinate (Mma in filter s and r dimension)
reset_bytes = reset_bytes_r_;
}
#else

2
include/cutlass/conv/threadblock/conv2d_wgrad_activation_tile_access_iterator_analytic.h
View File

@@ -195,7 +195,7 @@ public:
s = filter_s_[iteration_contiguous_];
}
else {
/// Multiple access to support non-128b alignment in contiguous dimenstion
/// Multiple access to support non-128b alignment in contiguous dimension
c = (filter_c_[iteration_contiguous_] + iteration_vector_ * AccessType::kElements) % problem_size_.C;
int wrap_c = (filter_c_[iteration_contiguous_] + iteration_vector_ * AccessType::kElements) / problem_size_.C;
s = (filter_s_[iteration_contiguous_] + wrap_c) % problem_size_.S;

2
include/cutlass/conv/threadblock/conv2d_wgrad_activation_tile_access_iterator_optimized.h
View File

@@ -212,7 +212,7 @@ public:
if (kAccessesPerVector > 1) {
// This code section is only to support non-128b alignment
// Multiple access to support non-128b alignment in contiguous dimenstion
// Multiple access to support non-128b alignment in contiguous dimension
int wrap_c;
params_.c_divmod(wrap_c, c, c + iteration_vector_ * AccessType::kElements);

2
include/cutlass/conv/threadblock/depthwise_fprop_pipelined.h
View File

@@ -241,7 +241,7 @@ public:
int rs_plane_idx = 0;
// Issue loads during the first warp-level matrix multiply-add *AFTER* issuing
// shared memory loads (which have the tighest latency requirement).
// shared memory loads (which have the tightest latency requirement).
//
// Mainloop

2
include/cutlass/conv/threadblock/implicit_gemm_pipelined.h
View File

@@ -238,7 +238,7 @@ public:
int smem_write_stage_idx = 1;
// Issue loads during the first warp-level matrix multiply-add *AFTER* issuing
// shared memory loads (which have the tighest latency requirement).
// shared memory loads (which have the tightest latency requirement).
//
// Mainloop

6
include/cutlass/conv/threadblock/threadblock_swizzle.h
View File

@@ -67,7 +67,7 @@ static int get_strided_dgrad_tile_m(
// CUTLASS strided dgrad performance for stride > filter, i.e., stride={2x2} and filter={1x1})
//
// * Optimization *
// Only launch CTAs in M dimenstion which contribute to a row in Dx output
// Only launch CTAs in M dimension which contribute to a row in Dx output
//
//
// * Constraints *
@@ -107,7 +107,7 @@ struct StridedDgradHorizontalThreadblockSwizzle :
// compute number of tiles in m dimension
int tile_m = get_strided_dgrad_tile_m(problem_size, tile_size.m());
// compute number of tiles in n dimenstion
// compute number of tiles in n dimension
int tile_n = (implicit_gemm_problem_size.n() + tile_size.n() - 1) / tile_size.n();
return gemm::GemmCoord(
@@ -148,7 +148,7 @@ struct StridedDgradIdentityThreadblockSwizzle :
// compute number of tiles in m dimension
int tile_m = get_strided_dgrad_tile_m(problem_size, tile_size.m());
// compute number of tiles in n dimenstion
// compute number of tiles in n dimension
int tile_n = (implicit_gemm_problem_size.n() + tile_size.n() - 1) / tile_size.n();
return gemm::GemmCoord(