Use range instead of single upper_bound

Author: lialan

mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td

@@ -216,10 +216,7 @@ def ROCDL_BlockIdXOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.x">;
 def ROCDL_BlockIdYOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.y">;
 def ROCDL_BlockIdZOp : ROCDL_SpecialIdRegisterOp<"workgroup.id.z">;
 
-def ROCDL_WavefrontSizeOp : ROCDL_IntrPure1Op<"wavefrontsize">,
-    Arguments<(ins OptionalAttr<IndexAttr>:$upper_bound)> {
-  let assemblyFormat = "(`upper_bound` $upper_bound^)? attr-dict `:` type($res)";
-}
+def ROCDL_WavefrontSizeOp : ROCDL_SpecialIdRegisterOp<"wavefrontsize">;
 
 //===----------------------------------------------------------------------===//
 // Thread range and Block range

mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp

@@ -137,8 +137,14 @@ struct GPUSubgroupSizeOpToROCDL : ConvertOpToLLVMPattern<gpu::SubgroupSizeOp> {
   LogicalResult
   matchAndRewrite(gpu::SubgroupSizeOp op, gpu::SubgroupSizeOp::Adaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
+    LLVM::ConstantRangeAttr bounds = nullptr;
+    if (auto upperBoundAttr = op.getUpperBoundAttr()) {
+      bounds = rewriter.getAttr<LLVM::ConstantRangeAttr>(
+          /*bitWidth=*/32, /*lower=*/32,
+          /*upper=*/op.getUpperBoundAttr().getInt());
+    }
     Value wavefrontOp = rewriter.create<ROCDL::WavefrontSizeOp>(
-        op.getLoc(), rewriter.getI32Type(), op.getUpperBoundAttr());
+        op.getLoc(), rewriter.getI32Type(), bounds);
     wavefrontOp = truncOrExtToLLVMType(rewriter, op.getLoc(), wavefrontOp,
                                        *getTypeConverter());
     rewriter.replaceOp(op, {wavefrontOp});

mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir

@@ -63,7 +63,7 @@ gpu.module @test_module {
     // CHECK: = llvm.sext %{{.*}} : i32 to i64
     %subgroupSize = gpu.subgroup_size : index
 
-    // CHECK: = rocdl.wavefrontsize upper_bound 64 : i32
+    // CHECK: = rocdl.wavefrontsize range <i32, 32, 64> : i32
     // CHECK: = llvm.sext %{{.*}} : i32 to i64
     %subgroupSize2 = gpu.subgroup_size upper_bound 64 : index
 

mlir/test/Target/LLVMIR/rocdl.mlir

@@ -36,8 +36,8 @@ llvm.func @rocdl_special_regs() -> i32 {
   // CHECK: call i32 @llvm.amdgcn.wavefrontsize()
   %15 = rocdl.wavefrontsize : i32
 
-  // CHECK: call i32 @llvm.amdgcn.wavefrontsize()
-  %16 = rocdl.wavefrontsize upper_bound 32 : i32
+  // CHECK: call range(i32 32, 64) i32 @llvm.amdgcn.wavefrontsize()
+  %16 = rocdl.wavefrontsize range <i32, 32, 64> : i32
 
   llvm.return %1 : i32
 }