Skip to content

Commit

Permalink
scheduler comments
Browse files Browse the repository at this point in the history
  • Loading branch information
rdspring1 committed Nov 4, 2024
1 parent 53aeda4 commit f8aa777
Showing 1 changed file with 4 additions and 2 deletions.
6 changes: 4 additions & 2 deletions csrc/scheduler/hopper_multi_matmul.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -31,8 +31,10 @@ namespace nvfuser {

namespace {

// Get gcd between megabanks and row stride AND repeated pattern size.
// If gcd is 1, then there is no swizzle is necessary to resolve bank conflicts.
// This function returns a pair of integers. The first integer is the gcd
// between megabanks and row stride. The second integer is the repeat pattern
// size. If the gcd is 1, then no swizzle is necessary to resolve bank
// conflicts. In that case, the second integer is irrelevant and -1 is returned.
std::pair<int64_t, int64_t> analyzeSwizzleSharedMemory(
TensorView* shared_mem_tv) {
NVF_ERROR(shared_mem_tv->getMemoryType() == MemoryType::Shared);
Expand Down

0 comments on commit f8aa777

Please sign in to comment.