[llvm-branch-commits] [mlir] [mlir][GPU] Plumb range information through the NVVM lowterings (PR #107659)

[Guray Ozen via llvm-branch-commits]

================
@@ -209,7 +209,12 @@ struct GPULaneIdOpToNVVM : ConvertOpToLLVMPattern<gpu::LaneIdOp> {
                   ConversionPatternRewriter &rewriter) const override {
     auto loc = op->getLoc();
     MLIRContext *context = rewriter.getContext();
-    Value newOp = rewriter.create<NVVM::LaneIdOp>(loc, rewriter.getI32Type());
+    LLVM::ConstantRangeAttr bounds = nullptr;
+    if (std::optional<APInt> upperBound = op.getUpperBound())
+      bounds = rewriter.getAttr<LLVM::ConstantRangeAttr>(
+          /*bitWidth=*/32, /*lower=*/0, upperBound->getZExtValue());
+    Value newOp =
+        rewriter.create<NVVM::LaneIdOp>(loc, rewriter.getI32Type(), bounds);
----------------
grypp wrote:

Can we always use `kWarpSize = 32` for the laneId? This is a HW constraint, and it hasn't been changed over the years. 

https://github.com/llvm/llvm-project/pull/107659