[Mlir-commits] [mlir] [MLIR][Math][XeVM] Add MathToXeVM (`math-to-xevm`) pass (PR #159878)

[Adam Siemieniuk]

================
@@ -796,6 +796,32 @@ def ConvertMathToSPIRVPass : Pass<"convert-math-to-spirv"> {
   let dependentDialects = ["spirv::SPIRVDialect"];
 }
 
+//===----------------------------------------------------------------------===//
+// MathToXeVM
+//===----------------------------------------------------------------------===//
+
+def ConvertMathToXeVM : Pass<"convert-math-to-xevm", "ModuleOp"> {
+  let summary =
+      "Convert (fast) math operations to native XeVM/SPIRV equivalents";
+  let description = [{
+    This pass converts supported math ops marked with the `afn` fastmath flag
+    to function calls for OpenCL `native_` math intrinsics: These intrinsics
+    are typically mapped directly to native device instructions, often resulting
+    in better performance. However, the precision/error of these intrinsics
+    are implementation-defined, and thus math ops are only converted when they
+    have the `afn` fastmath flag enabled.
+  }];
+  let options = [Option<
+      "convertArith", "convert-arith", "bool", /*default=*/"true",
+      "Convert supported Arith ops (e.g. arith.divf) as well.">];
+  let dependentDialects = [
+    "arith::ArithDialect",
+    "func::FuncDialect",
+    "xevm::XeVMDialect",
+    "vector::VectorDialect",
----------------
adam-smnk wrote:

I don't think func and vector dialects are needed here (and in the legal dialects registration). AFAIK, there's no creation of ops from these dialects.

Instead, llvm dialect is missing for the intrinsic calls.

https://github.com/llvm/llvm-project/pull/159878