cleanup ir, make it valid

Author: ZuseZ4

compiler/rustc_codegen_llvm/src/builder/gpu_offload.rs

@@ -354,32 +354,29 @@ fn gen_call_handling<'ll>(
     //%kernel_args = alloca %struct.__tgt_kernel_arguments, align 8
     let a5 = builder.direct_alloca(tgt_kernel_decl, Align::EIGHT, "kernel_args");
 
+    // Step 1)
+    unsafe { llvm::LLVMRustPositionBefore(builder.llbuilder, kernel_call) };
+    builder.memset(tgt_bin_desc_alloca, cx.get_const_i8(0), cx.get_const_i64(32), Align::EIGHT);
+
     // Now we allocate once per function param, a copy to be passed to one of our maps.
     let mut vals = vec![];
     let mut geps = vec![];
     let i32_0 = cx.get_const_i32(0);
     for (index, in_ty) in types.iter().enumerate() {
         // get function arg, store it into the alloca, and read it.
-        let p = llvm::get_param(called, index as u32);
-        let name = llvm::get_value_name(p);
-        let name = str::from_utf8(&name).unwrap();
-        let arg_name = format!("{name}.addr");
-        let alloca = builder.direct_alloca(in_ty, Align::EIGHT, &arg_name);
+        //let p = llvm::get_param(called, index as u32);
+        //let name = llvm::get_value_name(p);
+        //let name = str::from_utf8(&name).unwrap();
+        //let arg_name = format!("{name}.addr");
+        //let alloca = builder.direct_alloca(in_ty, Align::EIGHT, &arg_name);
 
         let v = unsafe { llvm::LLVMGetOperand(kernel_call, index as u32).unwrap() };
-        builder.store(v, alloca, Align::EIGHT);
-        //let val = builder.load(in_ty, alloca, Align::EIGHT);
-        //let gep = builder.inbounds_gep(cx.type_f32(), val, &[i32_0]);
         let gep = builder.inbounds_gep(cx.type_f32(), v, &[i32_0]);
         vals.push(v);
         //vals.push(val);
         geps.push(gep);
     }
 
-    // Step 1)
-    unsafe { llvm::LLVMRustPositionBefore(builder.llbuilder, kernel_call) };
-    builder.memset(tgt_bin_desc_alloca, cx.get_const_i8(0), cx.get_const_i64(32), Align::EIGHT);
-
     let mapper_fn_ty = cx.type_func(&[cx.type_ptr()], cx.type_void());
     let register_lib_decl = declare_offload_fn(&cx, "__tgt_register_lib", mapper_fn_ty);
     let unregister_lib_decl = declare_offload_fn(&cx, "__tgt_unregister_lib", mapper_fn_ty);