additional formatting and doc cleanups

Author: alexbaden

third_party/intel/include/Dialect/TritonIntelGPU/IR/TritonIntelGPUAttrDefs.td

@@ -300,9 +300,8 @@ def Subgroup2DBlockEncodingAttr : DistributedEncoding<"Subgroup2DBlockEncoding",
     - `threadsPerWarp` : currently a scalar, this parameter allows us to support different subgroup / warp configurations. Because the 2d block io operation is a subgroup operation, the size of the subgroup is important in determining the ordering of the loaded tensor.
     - `warpsPerCTA` : the number of warps per block / subgroups per workgroup and their distribution
     - `order` : The order within the block, used to determine along which dimension to broadcast.
-    - `kWidth` : used?
-    - `numReps` : unused?
-    - `CTALayout` : ??
+    - `kWidth` : Currently unused, but keeping because we will likely need it for layout conversions.
+    - `CTALayout` : Describes how blocks are distributed among work-groups/thread blocks.
   }];
 
   let parameters = (

third_party/intel/lib/Dialect/TritonIntelGPU/IR/Dialect.cpp

@@ -531,8 +531,8 @@ void maybePrintCTALayout(mlir::MLIRContext *context, mlir::AsmPrinter &printer,
 LogicalResult Subgroup2DBlockEncodingAttr::verify(
     function_ref<InFlightDiagnostic()> emitError,
     ArrayRef<unsigned> warpsPerCTA, CTALayoutAttr CTALayout,
-    ArrayRef<unsigned> instrShape, unsigned numBlocks, ArrayRef<unsigned> order, unsigned kWidth,
-    unsigned threadsPerWarp) {
+    ArrayRef<unsigned> instrShape, unsigned numBlocks, ArrayRef<unsigned> order,
+    unsigned kWidth, unsigned threadsPerWarp) {
   if (instrShape.size() != 2) {
     return emitError() << "instrShape must be rank 2 but was: "
                        << instrShape.size();
@@ -652,7 +652,8 @@ void Subgroup2DBlockEncodingAttr::print(AsmPrinter &printer) const {
   maybePrintCTALayout(getContext(), printer, getCTALayout(), getRank());
 
   printer << ", instrShape = [" << getInstrShape()
-          << "], numBlocks=" << getNumBlocks() << ", order=[" << getOrder() << "], kWidth=" << getKWidth()
+          << "], numBlocks=" << getNumBlocks() << ", order=[" << getOrder()
+          << "], kWidth=" << getKWidth()
           << ", threadsPerWarp=" << getThreadsPerWarp() << "}>";
 }
 

unittest/Dialect/TritonIntelGPU/LinearLayoutConversionsTest.cpp

@@ -35,7 +35,8 @@ class LinearLayoutConversionsTest : public ::testing::Test {
         CTALayoutAttr::get(
             &ctx, dpasLayout.getCTAsPerCGA(), // TODO: add to DpasLayout?
             dpasLayout.getCTASplitNum(), dpasLayout.getCTAOrder()),
-        instrShape, numBlocks, getOrderForDotOperand(opIdx, /*rank*/ 2, /*kContig*/ true), kWidth,
+        instrShape, numBlocks,
+        getOrderForDotOperand(opIdx, /*rank*/ 2, /*kContig*/ true), kWidth,
         dpasLayout.getThreadsPerWarp());
     return layout;
   }