Improve the performance of simple array comparisons

3 files changed, 49 insertions, 13 deletions

diff --git a/src/checker.cpp b/src/checker.cpp
index f68e2ab15..ba13b2237 100644
--- a/src/checker.cpp
+++ b/src/checker.cpp
@@ -1613,6 +1613,8 @@ void generate_minimum_dependency_set(Checker *c, Entity *start) {
 
 		str_lit("umodti3"),
 		str_lit("udivti3"),
+
+		str_lit("memory_compare"),
 	};
 	for (isize i = 0; i < gb_count_of(required_runtime_entities); i++) {
 		add_dependency_to_set(c, scope_lookup(c->info.runtime_package->scope, required_runtime_entities[i]));
diff --git a/src/ir.cpp b/src/ir.cpp
index 9f95db524..1105eac29 100644
--- a/src/ir.cpp
+++ b/src/ir.cpp
@@ -4126,20 +4126,31 @@ irValue *ir_emit_comp(irProcedure *proc, TokenKind op_kind, irValue *left, irVal
 
 			return ir_emit_load(proc, val);
 		} else {
-			irValue *val = ir_add_local_generated(proc, t_bool, false);
-			ir_emit_store(proc, val, res);
-			auto loop_data = ir_loop_start(proc, count, t_i32);
-			{
-				irValue *i = loop_data.idx;
-				irValue *x = ir_emit_load(proc, ir_emit_array_ep(proc, lhs, i));
-				irValue *y = ir_emit_load(proc, ir_emit_array_ep(proc, rhs, i));
-				irValue *cmp = ir_emit_comp(proc, op_kind, x, y);
-				irValue *new_res = ir_emit_arith(proc, cmp_op, ir_emit_load(proc, val), cmp, t_bool);
-				ir_emit_store(proc, val, ir_emit_conv(proc, new_res, t_bool));
-			}
-			ir_loop_end(proc, loop_data);
+			if (is_type_simple_compare(tl) && (op_kind == Token_CmpEq || op_kind == Token_NotEq)) {
+				// TODO(bill): Test to see if this is actually faster!!!!
+				auto args = array_make<irValue *>(heap_allocator(), 3);
+				args[0] = ir_emit_conv(proc, lhs, t_rawptr);
+				args[1] = ir_emit_conv(proc, rhs, t_rawptr);
+				args[2] = ir_const_int(type_size_of(tl));
+				irValue *val = ir_emit_runtime_call(proc, "memory_compare", args);
+				irValue *res = ir_emit_comp(proc, op_kind, val, v_zero);
+				return ir_emit_conv(proc, res, t_bool);
+			} else {
+				irValue *val = ir_add_local_generated(proc, t_bool, false);
+				ir_emit_store(proc, val, res);
+				auto loop_data = ir_loop_start(proc, count, t_i32);
+				{
+					irValue *i = loop_data.idx;
+					irValue *x = ir_emit_load(proc, ir_emit_array_ep(proc, lhs, i));
+					irValue *y = ir_emit_load(proc, ir_emit_array_ep(proc, rhs, i));
+					irValue *cmp = ir_emit_comp(proc, op_kind, x, y);
+					irValue *new_res = ir_emit_arith(proc, cmp_op, ir_emit_load(proc, val), cmp, t_bool);
+					ir_emit_store(proc, val, ir_emit_conv(proc, new_res, t_bool));
+				}
+				ir_loop_end(proc, loop_data);
 
-			return ir_emit_load(proc, val);
+				return ir_emit_load(proc, val);
+			}
 		}
 	}
 
diff --git a/src/types.cpp b/src/types.cpp
index dc7ecb946..5cf86d6b6 100644
--- a/src/types.cpp
+++ b/src/types.cpp
@@ -1037,6 +1037,29 @@ Type *core_array_type(Type *t) {
 	return t;
 }
 
+// NOTE(bill): type can be easily compared using memcmp
+bool is_type_simple_compare(Type *t) {
+	t = core_type(t);
+	switch (t->kind) {
+	case Type_Array:
+		return is_type_simple_compare(t->Array.elem);
+
+	case Type_Basic:
+		if (t->Basic.flags & (BasicFlag_Integer|BasicFlag_Float|BasicFlag_Complex|BasicFlag_Rune|BasicFlag_Pointer)) {
+			return true;
+		}
+		return false;
+
+	case Type_Pointer:
+	case Type_Proc:
+	case Type_BitSet:
+	case Type_BitField:
+		return true;
+	}
+
+	return false;
+}
+
 Type *base_complex_elem_type(Type *t) {
 	t = core_type(t);
 	if (is_type_complex(t)) {