masahi commented on code in PR #10923:
URL: https://github.com/apache/tvm/pull/10923#discussion_r844580948
##########
src/tir/analysis/verify_gpu_code.cc:
##########
@@ -61,11 +62,12 @@ class GPUCodeVerifier : public StmtExprVisitor {
void VisitStmt_(const AllocateNode* op) final {
StmtVisitor::VisitStmt_(op);
auto scope = GetPtrStorageScope(op->buffer_var);
+ runtime::StorageScope storage_scope = runtime::StorageScope::Create(scope);
// visit an allocation of a buffer in shared memory, record its size
- if (scope == "local") {
+ if (storage_scope.rank == runtime::StorageRank::kLocal) {
size_t size = static_cast<size_t>(op->ConstantAllocationSize());
local_memory_per_block_ += size * op->dtype.bytes() * op->dtype.lanes();
- } else if (scope == "shared") {
+ } else if (storage_scope.rank == runtime::StorageRank::kShared) {
size_t size = static_cast<size_t>(op->ConstantAllocationSize());
Review Comment:
Note that `op->ConstantAllocationSize()` is 0 for symbolic shape dyn-shared
memory. But `VerifyGPUCode` cannot do anything with symbolic shape, so this
should be fine.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
