support arithmetic for non byte aligned integer types

see #261

1 files changed, 66 insertions, 57 deletions

diff --git a/src/codegen.cpp b/src/codegen.cpp
index 0d8ef0b03a..3d5f675787 100644
--- a/src/codegen.cpp
+++ b/src/codegen.cpp
@@ -62,6 +62,7 @@ CodeGen *codegen_create(Buf *root_source_dir, const ZigTarget *target) {
     g->fn_type_table.init(32);
     g->error_table.init(16);
     g->generic_table.init(16);
+    g->llvm_fn_table.init(16);
     g->memoized_fn_eval_table.init(16);
     g->is_release_build = false;
     g->is_test_build = false;
@@ -352,33 +353,14 @@ static void clear_debug_source_node(CodeGen *g) {
     ZigLLVMClearCurrentDebugLocation(g->builder);
 }
 
-enum AddSubMul {
-    AddSubMulAdd = 0,
-    AddSubMulSub = 1,
-    AddSubMulMul = 2,
-};
-
-static size_t bits_index(size_t size_in_bits) {
-    switch (size_in_bits) {
-        case 8:
-            return 0;
-        case 16:
-            return 1;
-        case 32:
-            return 2;
-        case 64:
-            return 3;
-        default:
-            zig_unreachable();
-    }
-}
-
 static LLVMValueRef get_arithmetic_overflow_fn(CodeGen *g, TypeTableEntry *type_entry,
         const char *signed_name, const char *unsigned_name)
 {
+    char fn_name[64];
+
     assert(type_entry->id == TypeTableEntryIdInt);
     const char *signed_str = type_entry->data.integral.is_signed ? signed_name : unsigned_name;
-    Buf *llvm_name = buf_sprintf("llvm.%s.with.overflow.i%zu", signed_str, type_entry->data.integral.bit_count);
+    sprintf(fn_name, "llvm.%s.with.overflow.i%zu", signed_str, type_entry->data.integral.bit_count);
 
     LLVMTypeRef return_elem_types[] = {
         type_entry->type_ref,
@@ -390,34 +372,39 @@ static LLVMValueRef get_arithmetic_overflow_fn(CodeGen *g, TypeTableEntry *type_
     };
     LLVMTypeRef return_struct_type = LLVMStructType(return_elem_types, 2, false);
     LLVMTypeRef fn_type = LLVMFunctionType(return_struct_type, param_types, 2, false);
-    LLVMValueRef fn_val = LLVMAddFunction(g->module, buf_ptr(llvm_name), fn_type);
+    LLVMValueRef fn_val = LLVMAddFunction(g->module, fn_name, fn_type);
     assert(LLVMGetIntrinsicID(fn_val));
     return fn_val;
 }
 
 static LLVMValueRef get_int_overflow_fn(CodeGen *g, TypeTableEntry *type_entry, AddSubMul add_sub_mul) {
     assert(type_entry->id == TypeTableEntryIdInt);
-    // [0-signed,1-unsigned][0-add,1-sub,2-mul][0-8,1-16,2-32,3-64]
-    size_t index0 = type_entry->data.integral.is_signed ? 0 : 1;
-    size_t index1 = add_sub_mul;
-    size_t index2 = bits_index(type_entry->data.integral.bit_count);
-    LLVMValueRef *fn = &g->int_overflow_fns[index0][index1][index2];
-    if (*fn) {
-        return *fn;
-    }
+
+    ZigLLVMFnKey key = {};
+    key.id = ZigLLVMFnIdOverflowArithmetic;
+    key.data.overflow_arithmetic.is_signed = type_entry->data.integral.is_signed;
+    key.data.overflow_arithmetic.add_sub_mul = add_sub_mul;
+    key.data.overflow_arithmetic.bit_count = type_entry->data.integral.bit_count;
+
+    auto existing_entry = g->llvm_fn_table.maybe_get(key);
+    if (existing_entry)
+        return existing_entry->value;
+
+    LLVMValueRef fn_val;
     switch (add_sub_mul) {
         case AddSubMulAdd:
-            *fn = get_arithmetic_overflow_fn(g, type_entry, "sadd", "uadd");
+            fn_val = get_arithmetic_overflow_fn(g, type_entry, "sadd", "uadd");
             break;
         case AddSubMulSub:
-            *fn = get_arithmetic_overflow_fn(g, type_entry, "ssub", "usub");
+            fn_val = get_arithmetic_overflow_fn(g, type_entry, "ssub", "usub");
             break;
         case AddSubMulMul:
-            *fn = get_arithmetic_overflow_fn(g, type_entry, "smul", "umul");
+            fn_val = get_arithmetic_overflow_fn(g, type_entry, "smul", "umul");
             break;
-
     }
-    return *fn;
+
+    g->llvm_fn_table.put(key, fn_val);
+    return fn_val;
 }
 
 static LLVMValueRef get_handle_value(CodeGen *g, LLVMValueRef ptr, TypeTableEntry *type, bool is_volatile) {
@@ -1388,13 +1375,22 @@ static LLVMValueRef ir_render_load_ptr(CodeGen *g, IrExecutable *executable, IrI
     bool is_volatile = ptr_type->data.pointer.is_volatile;
 
     uint32_t bit_offset = ptr_type->data.pointer.bit_offset;
-    if (bit_offset == 0)
-        return get_handle_value(g, ptr, child_type, is_volatile);
-
-    assert(!handle_is_ptr(child_type));
-
-    LLVMValueRef containing_int = LLVMBuildLoad(g->builder, ptr, "");
-    LLVMSetVolatile(containing_int, is_volatile);
+    LLVMValueRef containing_int;
+    if (bit_offset == 0) {
+        LLVMValueRef result_val = get_handle_value(g, ptr, child_type, is_volatile);
+        if (LLVMGetTypeKind(LLVMTypeOf(result_val)) == LLVMIntegerTypeKind &&
+            LLVMGetTypeKind(child_type->type_ref) == LLVMIntegerTypeKind &&
+            LLVMGetIntTypeWidth(child_type->type_ref) < LLVMGetIntTypeWidth(LLVMTypeOf(result_val)))
+        {
+            containing_int = result_val;
+        } else {
+            return result_val;
+        }
+    } else {
+        assert(!handle_is_ptr(child_type));
+        containing_int = LLVMBuildLoad(g->builder, ptr, "");
+        LLVMSetVolatile(containing_int, is_volatile);
+    }
 
     uint32_t child_bit_count = type_size_bits(g, child_type);
     uint32_t host_bit_count = LLVMGetIntTypeWidth(LLVMTypeOf(containing_int));
@@ -1748,21 +1744,34 @@ static LLVMValueRef ir_render_unwrap_maybe(CodeGen *g, IrExecutable *executable,
 }
 
 static LLVMValueRef get_int_builtin_fn(CodeGen *g, TypeTableEntry *int_type, BuiltinFnId fn_id) {
-    // [0-ctz,1-clz][0-8,1-16,2-32,3-64]
-    size_t index0 = (fn_id == BuiltinFnIdCtz) ? 0 : 1;
-    size_t index1 = bits_index(int_type->data.integral.bit_count);
-    LLVMValueRef *fn = &g->int_builtin_fns[index0][index1];
-    if (!*fn) {
-        const char *fn_name = (fn_id == BuiltinFnIdCtz) ? "cttz" : "ctlz";
-        Buf *llvm_name = buf_sprintf("llvm.%s.i%zu", fn_name, int_type->data.integral.bit_count);
-        LLVMTypeRef param_types[] = {
-            int_type->type_ref,
-            LLVMInt1Type(),
-        };
-        LLVMTypeRef fn_type = LLVMFunctionType(int_type->type_ref, param_types, 2, false);
-        *fn = LLVMAddFunction(g->module, buf_ptr(llvm_name), fn_type);
+    ZigLLVMFnKey key = {};
+    const char *fn_name;
+    if (fn_id == BuiltinFnIdCtz) {
+        fn_name = "cttz";
+        key.id = ZigLLVMFnIdCtz;
+        key.data.ctz.bit_count = int_type->data.integral.bit_count;
+    } else {
+        fn_name = "ctlz";
+        key.id = ZigLLVMFnIdClz;
+        key.data.clz.bit_count = int_type->data.integral.bit_count;
     }
-    return *fn;
+
+    auto existing_entry = g->llvm_fn_table.maybe_get(key);
+    if (existing_entry)
+        return existing_entry->value;
+
+    char llvm_name[64];
+    sprintf(llvm_name, "llvm.%s.i%zu", fn_name, int_type->data.integral.bit_count);
+    LLVMTypeRef param_types[] = {
+        int_type->type_ref,
+        LLVMInt1Type(),
+    };
+    LLVMTypeRef fn_type = LLVMFunctionType(int_type->type_ref, param_types, 2, false);
+    LLVMValueRef fn_val = LLVMAddFunction(g->module, llvm_name, fn_type);
+
+    g->llvm_fn_table.put(key, fn_val);
+
+    return fn_val;
 }
 
 static LLVMValueRef ir_render_clz(CodeGen *g, IrExecutable *executable, IrInstructionClz *instruction) {