[BACKEND] add dot conversion (mma version=2) (#672)

LLVM Conversion for Dot op.

Due to the lack of `convert_layout`, currently, the dot only supports
the following combination of operands

- `$a` in shared layout
- `$b` in shared layout
- `$c` in MMA layout(but only Splat-like, leaving the generic cases to
`convert_layout`)

This PR focus on `mma.16816` related logic support, leaving the other
cases to the following PR.

Co-authored-by: Philippe Tillet <phil@openai.com>

lib/Dialect/TritonGPU/IR/Dialect.cpp

@@ -177,9 +177,9 @@ unsigned SliceEncodingAttr::getElemsPerThread(ArrayRef<int64_t> shape) const {
 }
 
 unsigned MmaEncodingAttr::getElemsPerThread(ArrayRef<int64_t> shape) const {
-  // TODO:
-  assert(0 && "MmaEncodingAttr::getElemsPerThread not implemented");
-  return 0;
+  int threads = product(getWarpsPerCTA());
+  int numElem = product(shape);
+  return numElem / threads;
 }
 
 unsigned SharedEncodingAttr::getElemsPerThread(ArrayRef<int64_t> shape) const {