[Mlir-commits] [mlir] [mlir][AMDGPU] Add support for AMD f16 math library calls (PR #108809)

[Daniel Hernandez-Juarez]

================
@@ -39,16 +39,17 @@ template <typename OpTy>
 static void populateOpPatterns(LLVMTypeConverter &converter,
                                RewritePatternSet &patterns, StringRef f32Func,
                                StringRef f64Func,
+                               StringRef f16Func,
                                StringRef f32ApproxFunc = "") {
   patterns.add<ScalarizeVectorOpLowering<OpTy>>(converter);
   patterns.add<OpToFuncCallLowering<OpTy>>(converter, f32Func, f64Func,
-                                           f32ApproxFunc);
+                                           f32ApproxFunc, f16Func);
 }
 
 void mlir::populateMathToROCDLConversionPatterns(LLVMTypeConverter &converter,
                                                  RewritePatternSet &patterns) {
   // Handled by mathToLLVM: math::AbsIOp
-  // Handled by mathToLLVM: math::AbsFIOp
+  // Handled by mathToLLVM: math::AbsFOp
----------------
dhernandez0 wrote:

I'm not sure if we should handle fp16 math::AbsFOp and math::SqrtOp here.

https://github.com/llvm/llvm-project/pull/108809