Orion34C commented on a change in pull request #4546: [CODEGEN] Support cuda
tensorcore subbyte int data type in auto tensorcore
URL: https://github.com/apache/incubator-tvm/pull/4546#discussion_r362204403
##########
File path: src/codegen/codegen_cuda.cc
##########
@@ -401,8 +441,10 @@ void CodeGenCUDA::VisitStmt_(const Allocate* op) {
std::string scope = alloc_storage_scope_.at(buffer);
if (scope.find("wmma.") == 0) {
if (scope == "wmma.matrix_a" || scope == "wmma.matrix_b") {
- CHECK(op->type == Float(16) || op->type == Int(8) || op->type ==
UInt(8))
- << "Matrix_a and matrix_b only support half or char or unsigned char
type for now";
+ CHECK(op->type == Float(16) || op->type == Int(8) || op->type ==
UInt(8)
+ || op->type == Int(4) || op->type == Int(1))
Review comment:
done
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
