Handle intrinsics in a more efficent manner.

src/intrinsic/llvm.rs

@@ -1548,10 +1548,11 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function
         "llvm.x86.tcmmrlfp16ps" => "__builtin_trap",
 
         // NOTE: this file is generated by https://github.com/GuillaumeGomez/llvmint/blob/master/generate_list.py
-        _ => include!("archs.rs"),
+        _ => map_arch_intrinsic(name),
     };
 
     let func = cx.context.get_target_builtin_function(gcc_name);
     cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
     func
 }
+include!("archs.rs");

tools/generate_intrinsics.py

@@ -168,25 +168,39 @@ def update_intrinsics(llvm_path, llvmint, llvmint2):
         os.path.dirname(os.path.abspath(__file__)),
         "../src/intrinsic/archs.rs",
     )
+    # A hashmap of all architectures. This allows us to first match on the architecture, and then on the intrisnics. 
+    # This speeds up the comparison, and makes our code considerably smaller.
+    # Since all intrinsic names start with ".llvm", we skip that prefix. 
     print("Updating content of `{}`...".format(output_file))
     with open(output_file, "w", encoding="utf8") as out:
         out.write("// File generated by `rustc_codegen_gcc/tools/generate_intrinsics.py`\n")
         out.write("// DO NOT EDIT IT!\n")
-        out.write("match name {\n")
+        out.write("/// Translate a given LLVM intrinsic name to an equivalent GCC one.\n")
+        out.write("fn map_arch_intrinsic(name:&str)->&str{\n")
+        out.write('let Some(name) =  name.strip_prefix("llvm.") else {unimplemented!("***** unsupported LLVM intrinsic {}", name)};\n')
+        out.write('let Some((arch, name)) =  name.split_once(\'.\') else {unimplemented!("***** unsupported LLVM intrinsic {}", name)};\n')
+        out.write("match arch {\n")
         for arch in archs:
             if len(intrinsics[arch]) == 0:
                 continue
+            out.write("\"{}\" => {{ #[allow(non_snake_case)] fn {}(name:&str)->&str{{ match name{{".format(arch,arch))
             intrinsics[arch].sort(key=lambda x: (x[0], x[2]))
             out.write('    // {}\n'.format(arch))
             for entry in intrinsics[arch]:
+                llvm_name = entry[0].removeprefix("llvm.");
+                llvm_name = llvm_name.removeprefix(arch);
+                llvm_name = llvm_name.removeprefix(".");
                 if entry[2] is True: # if it is a duplicate
-                    out.write('    // [DUPLICATE]: "{}" => "{}",\n'.format(entry[0], entry[1]))
+                    out.write('    // [DUPLICATE]: "{}" => "{}",\n'.format(llvm_name, entry[1]))
                 elif "_round_mask" in entry[1]:
-                    out.write('    // [INVALID CONVERSION]: "{}" => "{}",\n'.format(entry[0], entry[1]))
+                    out.write('    // [INVALID CONVERSION]: "{}" => "{}",\n'.format(llvm_name, entry[1]))
                 else:
-                    out.write('    "{}" => "{}",\n'.format(entry[0], entry[1]))
+                    out.write('    "{}" => "{}",\n'.format(llvm_name, entry[1]))
+            out.write('    _ => unimplemented!("***** unsupported LLVM intrinsic {}", name),\n')
+            out.write("}} }} {}(name) }}\n,".format(arch))
         out.write('    _ => unimplemented!("***** unsupported LLVM intrinsic {}", name),\n')
-        out.write("}\n")
+        out.write("}\n}")
+    subprocess.call(["rustfmt", output_file])
     print("Done!")