[Mlir-commits] [mlir] [mlir][Math][LLVMSPV] Add Math to LLVM-SPV conversion pass (PR #198370)

[Md Abdullah Shahneous Bari]

================
@@ -119,6 +123,76 @@ struct ConvertNativeFuncPattern final : public OpConversionPattern<Op> {
   const StringRef nativeFunc;
 };
 
+template <typename OpTy>
+static void populateOCLExtSetOpPatterns(const LLVMTypeConverter &converter,
+                                        RewritePatternSet &patterns,
+                                        PatternBenefit benefit,
+                                        StringRef opName) {
+  std::string prefix = "__spirv_ocl_";
+  std::string mangledName = "_Z" +
+                            std::to_string(prefix.size() + opName.size()) +
+                            prefix + opName.str();
+
----------------
mshahneo wrote:

For SPIR-V targeted vector lowering, the vector size should be SPIR-V spec compliant (2, 3, 4) and (8, 16 if vector16 Capability is enabled). I think it's better to check for it.

There is already a utility in this file `isSPIRVCompatibleFloatOrVec`: https://github.com/llvm/llvm-project/blob/main/mlir/lib/Conversion/MathToXeVM/MathToXeVM.cpp#L75

You can use that.

https://github.com/llvm/llvm-project/pull/198370