[DOCS] Add comments in layout.h (#249)

include/triton/codegen/analysis/layout.h

@@ -111,10 +111,15 @@ public:
   int rep(size_t k) { return rep_.at(k); }
 
 private:
+  // fragment per warp
   std::vector<int> fpw_;
+  // shape per warp
   std::vector<int> spw_;
+  // warp per tile
   std::vector<int> wpt_;
+  // shape per tile
   std::vector<int> spt_;
+  // repetitions
   std::vector<int> rep_;
 };
 
@@ -131,7 +136,9 @@ struct scanline_layout: public data_layout {
   int nts(size_t k) { return nts_.at(k); }
 
 public:
+  // micro tile size. The size of a tile held by a thread block.
   std::vector<int> mts_;
+  // nano tile size. The size of a tile held by a thread.
   std::vector<int> nts_;
 };
 

lib/codegen/selection/generator.cc

@@ -465,7 +465,7 @@ Value* generator::bf16_to_fp32(Value *in0){
 
 Value* generator::fp32_to_bf16(Value *in0){
   if(tgt_->as_nvidia()->sm() >= 80){
-    InlineAsm *ptx = InlineAsm::get(FunctionType::get(builder_->getInt16Ty(), {builder_->getFloatTy()}),
+    InlineAsm *ptx = InlineAsm::get(FunctionType::get(builder_->getInt16Ty(), {builder_->getFloatTy()}, false),
                                     "cvt.rn.bf16.f32 $0, $1;", "=h,r", false);
     return call(ptx, {in0});
   }

python/src/triton.cc

@@ -15,6 +15,7 @@
 #include <pybind11/stl.h>
 #include <regex>
 #include <string>
+#include <sstream>
 
 namespace py = pybind11;
 namespace ir = triton::ir;