huochaitiantang commented on a change in pull request #5226: [CODEGEN][CUDA]
Fix vector load
URL: https://github.com/apache/incubator-tvm/pull/5226#discussion_r405325498
##########
File path: src/target/source/codegen_cuda.cc
##########
@@ -591,13 +591,17 @@ void CodeGenCUDA::VisitExpr_(const RampNode* op,
std::ostream& os) {
}
void CodeGenCUDA::VisitExpr_(const BroadcastNode* op, std::ostream& os) { //
NOLINT(*)
- if (op->dtype.is_int() && op->dtype.bits() == 8 && op->lanes == 4) {
+ if ((op->dtype.is_int() || op->dtype.is_uint()) && op->dtype.bits() == 8 &&
op->lanes == 4) {
// make_int8x4
const int64_t *p = as_const_int(op->value);
CHECK(p);
int64_t v = *p & 0xFF;
v = (v << 24) | (v << 16) | (v << 8) | v;
- os << "(int)" << v;
+ if (op->dtype.is_uint()) {
Review comment:
TVM uses uint to store unit8x4 (in function PrintType). The care will
generate code like `unit x = (unit)y`, instead of `unit x = (int)y`. And what
is your further opinion?
----------------------------------------------------------------
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:
us...@infra.apache.org
With regards,
Apache Git Services
