diff --git a/lib/std/core/mem.c3 b/lib/std/core/mem.c3
index 398db9f68..51d0eaa56 100644
--- a/lib/std/core/mem.c3
+++ b/lib/std/core/mem.c3
@@ -26,7 +26,6 @@ macro void* aligned_pointer(void* ptr, usz alignment)
 	return (void*)(uptr)aligned_offset((uptr)ptr, alignment);
 }
 
-
 /**
  * @require math::is_power_of_2(alignment)
  **/
@@ -35,14 +34,22 @@ fn bool ptr_is_aligned(void* ptr, usz alignment) @inline
 	return (uptr)ptr & ((uptr)alignment - 1) == 0;
 }
 
-macro void copy(void* dst, void* src, usz len, usz $dst_align = 0, usz $src_align = 0, bool $is_volatile = false)
+macro void clear(void* dst, usz len, usz $dst_align = 0, bool $is_volatile = false, bool $inlined = false)
 {
-	$$memcpy(dst, src, len, $is_volatile, $dst_align, $src_align);
+	$if ($inlined):
+		$$memset_inline(dst, (char)0, len, $is_volatile, $dst_align);
+	$else:
+		$$memset(dst, (char)0, len, $is_volatile, $dst_align);
+	$endif;
 }
 
-macro void set(void* dst, char val, usz len, usz $dst_align = 0, bool $is_volatile = false)
+macro void copy(void* dst, void* src, usz len, usz $dst_align = 0, usz $src_align = 0, bool $is_volatile = false, bool $inlined = false)
 {
-	$$memset(dst, val, len, $is_volatile, $dst_align);
+	$if ($inlined):
+		$$memcpy_inline(dst, src, len, $is_volatile, $dst_align, $src_align);
+	$else:
+		$$memcpy(dst, src, len, $is_volatile, $dst_align, $src_align);
+	$endif;
 }
 
 macro void move(void* dst, void* src, usz len, usz $dst_align = 0, usz $src_align = 0, bool $is_volatile = false)
@@ -50,9 +57,13 @@ macro void move(void* dst, void* src, usz len, usz $dst_align = 0, usz $src_alig
 	$$memmove(dst, src, len, $is_volatile, $dst_align, $src_align);
 }
 
-macro void clear(void* dst, usz len, usz $dst_align = 0, bool $is_volatile = false)
+macro void set(void* dst, char val, usz len, usz $dst_align = 0, bool $is_volatile = false, bool $inlined = false)
 {
-	$$memset(dst, (char)0, len, $is_volatile, $dst_align);
+	$if ($inlined):
+		$$memset_inline(dst, val, len, $is_volatile, $dst_align);
+	$else:
+		$$memset(dst, val, len, $is_volatile, $dst_align);
+	$endif;
 }
 
 /**
diff --git a/src/compiler/enums.h b/src/compiler/enums.h
index 2c1689b90..052e04f9f 100644
--- a/src/compiler/enums.h
+++ b/src/compiler/enums.h
@@ -850,8 +850,10 @@ typedef enum
 	BUILTIN_LOG2,
 	BUILTIN_MAX,
 	BUILTIN_MEMCOPY,
+	BUILTIN_MEMCOPY_INLINE,
 	BUILTIN_MEMMOVE,
 	BUILTIN_MEMSET,
+	BUILTIN_MEMSET_INLINE,
 	BUILTIN_MIN,
 	BUILTIN_NEARBYINT,
 	BUILTIN_OVERFLOW_ADD,
diff --git a/src/compiler/llvm_codegen.c b/src/compiler/llvm_codegen.c
index 53dfd6fa3..1ea946b06 100644
--- a/src/compiler/llvm_codegen.c
+++ b/src/compiler/llvm_codegen.c
@@ -658,8 +658,10 @@ static void llvm_codegen_setup()
 	intrinsic_id.maximum = lookup_intrinsic("llvm.maximum");
 	intrinsic_id.maxnum = lookup_intrinsic("llvm.maxnum");
 	intrinsic_id.memcpy = lookup_intrinsic("llvm.memcpy");
-	intrinsic_id.memset = lookup_intrinsic("llvm.memset");
+	intrinsic_id.memcpy_inline = lookup_intrinsic("llvm.memcpy.inline");
 	intrinsic_id.memmove = lookup_intrinsic("llvm.memmove");
+	intrinsic_id.memset = lookup_intrinsic("llvm.memset");
+	intrinsic_id.memset_inline = lookup_intrinsic("llvm.memset.inline");
 	intrinsic_id.minimum = lookup_intrinsic("llvm.minimum");
 	intrinsic_id.minnum = lookup_intrinsic("llvm.minnum");
 	intrinsic_id.fmuladd = lookup_intrinsic("llvm.fmuladd");
diff --git a/src/compiler/llvm_codegen_builtins.c b/src/compiler/llvm_codegen_builtins.c
index 169dc0431..82e3ffcd9 100644
--- a/src/compiler/llvm_codegen_builtins.c
+++ b/src/compiler/llvm_codegen_builtins.c
@@ -215,7 +215,7 @@ INLINE void llvm_emit_intrinsic_args(GenContext *c, Expr **args, LLVMValueRef *s
 	}
 }
 
-INLINE void llvm_emit_memcpy_builtin(GenContext *c, BEValue *be_value, Expr *expr)
+INLINE void llvm_emit_memcpy_builtin(GenContext *c, unsigned intrinsic, BEValue *be_value, Expr *expr)
 {
 	Expr **args = expr->call_expr.arguments;
 	LLVMValueRef arg_slots[4];
@@ -225,7 +225,7 @@ INLINE void llvm_emit_memcpy_builtin(GenContext *c, BEValue *be_value, Expr *exp
 	LLVMTypeRef call_type[3];
 	call_type[0] = call_type[1] = llvm_get_type(c, type_voidptr);
 	call_type[2] = llvm_get_type(c, type_usize);
-	LLVMValueRef result = llvm_emit_call_intrinsic(c, intrinsic_id.memcpy, call_type, 3, arg_slots, 4);
+	LLVMValueRef result = llvm_emit_call_intrinsic(c, intrinsic, call_type, 3, arg_slots, 4);
 	assert(args[4]->const_expr.const_kind == CONST_INTEGER);
 	assert(args[5]->const_expr.const_kind == CONST_INTEGER);
 	uint64_t dst_align = int_to_u64(args[4]->const_expr.ixx);
@@ -235,20 +235,6 @@ INLINE void llvm_emit_memcpy_builtin(GenContext *c, BEValue *be_value, Expr *exp
 	llvm_value_set(be_value, result, type_void);
 }
 
-INLINE void llvm_emit_memset_builtin(GenContext *c, BEValue *be_value, Expr *expr)
-{
-	Expr **args = expr->call_expr.arguments;
-	LLVMValueRef arg_slots[4];
-	llvm_emit_intrinsic_args(c, args, arg_slots, 4);
-	arg_slots[0] = llvm_emit_bitcast(c, arg_slots[0], type_voidptr);
-	LLVMTypeRef call_type[2] = { llvm_get_type(c, type_voidptr), llvm_get_type(c, type_usize) };
-	LLVMValueRef result = llvm_emit_call_intrinsic(c, intrinsic_id.memset, call_type, 2, arg_slots, 4);
-	assert(args[4]->const_expr.const_kind == CONST_INTEGER);
-	uint64_t dst_align = int_to_u64(args[4]->const_expr.ixx);
-	if (dst_align > 0) llvm_attribute_add_call(c, result, attribute_id.align, 1, dst_align);
-	llvm_value_set(be_value, result, type_void);
-}
-
 INLINE void llvm_emit_memmove_builtin(GenContext *c, BEValue *be_value, Expr *expr)
 {
 	Expr **args = expr->call_expr.arguments;
@@ -268,6 +254,21 @@ INLINE void llvm_emit_memmove_builtin(GenContext *c, BEValue *be_value, Expr *ex
 	if (src_align > 0) llvm_attribute_add_call(c, result, attribute_id.align, 2, src_align);
 	llvm_value_set(be_value, result, type_void);
 }
+
+INLINE void llvm_emit_memset_builtin(GenContext *c, unsigned intrinsic, BEValue *be_value, Expr *expr)
+{
+	Expr **args = expr->call_expr.arguments;
+	LLVMValueRef arg_slots[4];
+	llvm_emit_intrinsic_args(c, args, arg_slots, 4);
+	arg_slots[0] = llvm_emit_bitcast(c, arg_slots[0], type_voidptr);
+	LLVMTypeRef call_type[2] = { llvm_get_type(c, type_voidptr), llvm_get_type(c, type_usize) };
+	LLVMValueRef result = llvm_emit_call_intrinsic(c, intrinsic, call_type, 2, arg_slots, 4);
+	assert(args[4]->const_expr.const_kind == CONST_INTEGER);
+	uint64_t dst_align = int_to_u64(args[4]->const_expr.ixx);
+	if (dst_align > 0) llvm_attribute_add_call(c, result, attribute_id.align, 1, dst_align);
+	llvm_value_set(be_value, result, type_void);
+}
+
 INLINE void llvm_emit_prefetch(GenContext *c, BEValue *be_value, Expr *expr)
 {
 	Expr **args = expr->call_expr.arguments;
@@ -473,14 +474,20 @@ void llvm_emit_builtin_call(GenContext *c, BEValue *result_value, Expr *expr)
 			llvm_emit_syscall(c, result_value, expr);
 			return;
 		case BUILTIN_MEMCOPY:
-			llvm_emit_memcpy_builtin(c, result_value, expr);
+			llvm_emit_memcpy_builtin(c, intrinsic_id.memcpy, result_value, expr);
 			return;
-		case BUILTIN_MEMSET:
-			llvm_emit_memset_builtin(c, result_value, expr);
+		case BUILTIN_MEMCOPY_INLINE:
+			llvm_emit_memcpy_builtin(c, intrinsic_id.memcpy_inline, result_value, expr);
 			return;
 		case BUILTIN_MEMMOVE:
 			llvm_emit_memmove_builtin(c, result_value, expr);
 			return;
+		case BUILTIN_MEMSET:
+			llvm_emit_memset_builtin(c, intrinsic_id.memset, result_value, expr);
+			return;
+		case BUILTIN_MEMSET_INLINE:
+			llvm_emit_memset_builtin(c, intrinsic_id.memset_inline, result_value, expr);
+			return;
 		case BUILTIN_SYSCLOCK:
 			llvm_value_set(result_value, llvm_emit_call_intrinsic(c, intrinsic_id.readcyclecounter, NULL, 0, NULL, 0), expr->type);
 			return;
diff --git a/src/compiler/llvm_codegen_internal.h b/src/compiler/llvm_codegen_internal.h
index 9f2477cab..96db89d81 100644
--- a/src/compiler/llvm_codegen_internal.h
+++ b/src/compiler/llvm_codegen_internal.h
@@ -134,6 +134,7 @@ typedef struct
 	unsigned floor;
 	unsigned flt_rounds;
 	unsigned fma;
+	unsigned fmuladd;
 	unsigned frameaddress;
 	unsigned fshl;
 	unsigned fshr;
@@ -149,11 +150,12 @@ typedef struct
 	unsigned maximum;
 	unsigned maxnum;
 	unsigned memcpy;
-	unsigned memset;
+	unsigned memcpy_inline;
 	unsigned memmove;
+	unsigned memset;
+	unsigned memset_inline;
 	unsigned minimum;
 	unsigned minnum;
-	unsigned fmuladd;
 	unsigned nearbyint;
 	unsigned pow;
 	unsigned powi;
diff --git a/src/compiler/sema_builtins.c b/src/compiler/sema_builtins.c
index 9773212dc..adbf856cc 100644
--- a/src/compiler/sema_builtins.c
+++ b/src/compiler/sema_builtins.c
@@ -349,6 +349,7 @@ bool sema_expr_analyse_builtin_call(SemaContext *context, Expr *expr)
 			rtype = type_no_optional(args[0]->type->canonical);
 			break;
 		case BUILTIN_MEMCOPY:
+		case BUILTIN_MEMCOPY_INLINE:
 		case BUILTIN_MEMMOVE:
 			if (!sema_check_builtin_args(args,
 										 (BuiltinArg[]) { BA_POINTER, BA_POINTER, BA_SIZE, BA_BOOL, BA_SIZE, BA_SIZE },
@@ -357,6 +358,7 @@ bool sema_expr_analyse_builtin_call(SemaContext *context, Expr *expr)
 			rtype = type_void;
 			break;
 		case BUILTIN_MEMSET:
+		case BUILTIN_MEMSET_INLINE:
 			if (!sema_check_builtin_args(args,
 										 (BuiltinArg[]) { BA_POINTER, BA_CHAR, BA_SIZE, BA_BOOL, BA_SIZE },
 										 arg_count)) return false;
@@ -620,11 +622,13 @@ static inline unsigned builtin_expected_args(BuiltinFunction func)
 		case BUILTIN_OVERFLOW_SUB:
 		case BUILTIN_PREFETCH:
 			return 3;
-		case BUILTIN_MEMSET:
-			return 5;
 		case BUILTIN_MEMCOPY:
+		case BUILTIN_MEMCOPY_INLINE:
 		case BUILTIN_MEMMOVE:
 			return 6;
+		case BUILTIN_MEMSET:
+		case BUILTIN_MEMSET_INLINE:
+			return 5;
 		case BUILTIN_SHUFFLEVECTOR:
 		case BUILTIN_NONE:
 			UNREACHABLE
diff --git a/src/compiler/symtab.c b/src/compiler/symtab.c
index b3ef2a6d1..21fa69f69 100644
--- a/src/compiler/symtab.c
+++ b/src/compiler/symtab.c
@@ -211,8 +211,10 @@ void symtab_init(uint32_t capacity)
 	builtin_list[BUILTIN_LOG2] = KW_DEF("log2");
 	builtin_list[BUILTIN_LOG10] = KW_DEF("log10");
 	builtin_list[BUILTIN_MEMCOPY] = KW_DEF("memcpy");
-	builtin_list[BUILTIN_MEMSET] = KW_DEF("memset");
+	builtin_list[BUILTIN_MEMCOPY_INLINE] = KW_DEF("memcpy_inline");
 	builtin_list[BUILTIN_MEMMOVE] = KW_DEF("memmove");
+	builtin_list[BUILTIN_MEMSET] = KW_DEF("memset");
+	builtin_list[BUILTIN_MEMSET_INLINE] = KW_DEF("memset_inline");
 	builtin_list[BUILTIN_NEARBYINT] = KW_DEF("nearbyint");
 	builtin_list[BUILTIN_OVERFLOW_ADD] = KW_DEF("overflow_add");
 	builtin_list[BUILTIN_OVERFLOW_SUB] = KW_DEF("overflow_sub");
diff --git a/test/test_suite/builtins/mem.c3t b/test/test_suite/builtins/mem.c3t
index 3770931f0..416fd4196 100644
--- a/test/test_suite/builtins/mem.c3t
+++ b/test/test_suite/builtins/mem.c3t
@@ -9,10 +9,14 @@ fn void main()
     $$memcpy(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
     $$memmove(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
     $$memset(&dst, 0u8, ushort.sizeof, false, $alignof(dst));
+
+    $$memcpy_inline(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
+    $$memset_inline(&dst, 0u8, ushort.sizeof, false, $alignof(dst));
 }
 
 /* #expect: test.ll
 
   call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 2 %0, i8* align 2 %1, i64 2, i1 false)
   call void @llvm.memmove.p0i8.p0i8.i64(i8* align 2 %2, i8* align 2 %3, i64 2, i1 false)
-  call void @llvm.memset.p0i8.i64(i8* align 2 %4, i8 0, i64 2, i1 false)
\ No newline at end of file
+  call void @llvm.memset.p0i8.i64(i8* align 2 %4, i8 0, i64 2, i1 false)
+  call void @llvm.memcpy.inline.p0i8.p0i8.i64(i8* align 2 %5, i8* align 2 %6, i64 2, i1 false)
diff --git a/test/test_suite/stdlib/map.c3t b/test/test_suite/stdlib/map.c3t
index 9863ce5b7..2cf149c0c 100644
--- a/test/test_suite/stdlib/map.c3t
+++ b/test/test_suite/stdlib/map.c3t
@@ -508,7 +508,7 @@ after_check100:                                   ; preds = %if.then
 
 panic_block:                                      ; preds = %assign_optional
   %160 = load void (i8*, i64, i8*, i64, i8*, i64, i32)*, void (i8*, i64, i8*, i64, i8*, i64, i32)** @std_core_builtin_panic, align 8
-  call void %160(i8* getelementptr inbounds ([28 x i8], [28 x i8]* @.panic_msg, i64 0, i64 0), i64 27, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @.file, i64 0, i64 0), i64 6, i8* getelementptr inbounds ([5 x i8], [5 x i8]* @.func, i64 0, i64 0), i64 4, i32 246)
+  call void %160(i8* getelementptr inbounds ([28 x i8], [28 x i8]* @.panic_msg, i64 0, i64 0), i64 27, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @.file, i64 0, i64 0), i64 6, i8* getelementptr inbounds ([5 x i8], [5 x i8]* @.func, i64 0, i64 0), i64 4, i32 257)
   unreachable
 
 noerr_block:                                      ; preds = %after_check100
diff --git a/test/test_suite2/builtins/mem.c3t b/test/test_suite2/builtins/mem.c3t
index 2acaf19bb..08c1ea6e1 100644
--- a/test/test_suite2/builtins/mem.c3t
+++ b/test/test_suite2/builtins/mem.c3t
@@ -9,6 +9,9 @@ fn void main()
     $$memcpy(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
     $$memmove(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
     $$memset(&dst, 0u8, ushort.sizeof, false, $alignof(dst));
+
+    $$memcpy_inline(&dst, &src, ushort.sizeof, false, $alignof(dst), $alignof(src));
+    $$memset_inline(&dst, 0u8, ushort.sizeof, false, $alignof(dst));
 }
 
 /* #expect: test.ll
@@ -16,3 +19,5 @@ fn void main()
   call void @llvm.memcpy.p0.p0.i64(ptr align 2 %dst, ptr align 2 %src, i64 2, i1 false)
   call void @llvm.memmove.p0.p0.i64(ptr align 2 %dst, ptr align 2 %src, i64 2, i1 false)
   call void @llvm.memset.p0.i64(ptr align 2 %dst, i8 0, i64 2, i1 false)
+  call void @llvm.memcpy.inline.p0.p0.i64(ptr align 2 %dst, ptr align 2 %src, i64 2, i1 false)
+  call void @llvm.memset.inline.p0.i64(ptr align 2 %dst, i8 0, i64 2, i1 false)
diff --git a/test/test_suite2/stdlib/map.c3t b/test/test_suite2/stdlib/map.c3t
index 77f9bc6cb..55c0e3a2b 100644
--- a/test/test_suite2/stdlib/map.c3t
+++ b/test/test_suite2/stdlib/map.c3t
@@ -364,7 +364,7 @@ after_check67:                                    ; preds = %if.then
 
 panic_block:                                      ; preds = %assign_optional
   %81 = load ptr, ptr @std_core_builtin_panic, align 8
-  call void %81(ptr @.panic_msg, i64 27, ptr @.file, i64 6, ptr @.func, i64 4, i32 246)
+  call void %81(ptr @.panic_msg, i64 27, ptr @.file, i64 6, ptr @.func, i64 4, i32 257)
   unreachable
 
 noerr_block:                                      ; preds = %after_check67