[Mlir-commits] [mlir] [MLIR][NVVM] Add sqrt Ops (PR #197422)

[Guray Ozen]

================
@@ -6544,6 +6544,41 @@ def NVVM_FmaOp : NVVM_Op<"fma", [Pure, SameOperandsAndResultType]> {
   }];
 }
 
+//===----------------------------------------------------------------------===//
+// NVVM sqrt op definitions
+//===----------------------------------------------------------------------===//
+
+def NVVM_SqrtOp
+    : NVVM_SingleResultIntrinsicOp<"sqrt", [Pure, SameOperandsAndResultType]> {
+  let summary = "Take the square root of a value";
+  let description = [{
+    Compute sqrt(src) and store the result in res.
+
+    For more information, see PTX ISA:
+    [sqrt](https://docs.nvidia.com/cuda/parallel-thread-execution/#floating-point-instructions-sqrt)
+  }];
+  let arguments = (ins AnyTypeOf<[F32, F64]>:$src, FPArithRoundingMode:$rnd,
+      DefaultValuedAttr<BoolAttr, "false">:$ftz);
+  let results = (outs AnyTypeOf<[F32, F64]>:$res);
+  let assemblyFormat = "$src attr-dict `:` type($src)";
+  let hasVerifier = 1;
+}
+
+def NVVM_SqrtApproxOp : NVVM_F32UnaryApproxOp<"sqrt.approx"> {
----------------
grypp wrote:

Do we really need extra OP for `approx`? 
@durga4github @schwarzschild-radius @varadk27 

https://github.com/llvm/llvm-project/pull/197422