Assorted fixes:

Fix creation of the target machine. Fix CPU features, architecture and target name.

Function names are not mandatory. Support unnamed functions.

Include parameters and local variables in the list of local variables for LLVMFunctionCodeGenerator.

Run the tests single-threaded, using a Mutex in the rayon code. This means that only the first error is real, subsequent errors are just picking up the panic across threads.

Add TODOs about attributes on vmctx.

lib/compiler-llvm/src/config.rs

@@ -100,8 +100,8 @@ impl LLVMConfig {
                 CpuFeature::SSE2 => Some("+sse2"),
                 CpuFeature::SSE3 => Some("+sse3"),
                 CpuFeature::SSSE3 => Some("+ssse3"),
-                CpuFeature::SSE41 => Some("+sse41"),
-                CpuFeature::SSE42 => Some("+sse42"),
+                CpuFeature::SSE41 => Some("+sse4.1"),
+                CpuFeature::SSE42 => Some("+sse4.2"),
                 CpuFeature::POPCNT => Some("+popcnt"),
                 CpuFeature::AVX => Some("+avx"),
                 CpuFeature::BMI1 => Some("+bmi"),
@@ -114,11 +114,11 @@ impl LLVMConfig {
             .join(",");
 
         let arch_string = triple.architecture.to_string();
-        let llvm_target = LLVMTarget::from_name(&arch_string).unwrap();
+        let llvm_target = LLVMTarget::from_triple(&self.target_triple()).unwrap();
         let target_machine = llvm_target
             .create_target_machine(
                 &self.target_triple(),
-                &arch_string,
+                "generic",
                 &llvm_cpu_features,
                 self.opt_level.clone(),
                 self.reloc_mode(),