py: Implement closures in native code generator.

Currently supports only x64 and Thumb2 archs.

py/asmx64.c

@@ -235,9 +235,15 @@ STATIC void asm_x64_write_word32_to(asm_x64_t *as, int offset, int w32) {
 */
 
 STATIC void asm_x64_write_r64_disp(asm_x64_t *as, int r64, int disp_r64, int disp_offset) {
-    assert(disp_r64 < 8);
     assert(disp_r64 != ASM_X64_REG_RSP);
 
+    if (disp_r64 == ASM_X64_REG_R12) {
+        // special case for r12; not fully implemented
+        assert(SIGNED_FIT8(disp_offset));
+        asm_x64_write_byte_3(as, MODRM_R64(r64) | MODRM_RM_DISP8 | MODRM_RM_R64(disp_r64), 0x24, IMM32_L0(disp_offset));
+        return;
+    }
+
     if (disp_offset == 0 && disp_r64 != ASM_X64_REG_RBP) {
         asm_x64_write_byte_1(as, MODRM_R64(r64) | MODRM_RM_DISP0 | MODRM_RM_R64(disp_r64));
     } else if (SIGNED_FIT8(disp_offset)) {
@@ -317,8 +323,7 @@ void asm_x64_mov_r16_to_mem16(asm_x64_t *as, int src_r64, int dest_r64, int dest
 
 void asm_x64_mov_r64_to_mem64(asm_x64_t *as, int src_r64, int dest_r64, int dest_disp) {
     // use REX prefix for 64 bit operation
-    assert(dest_r64 < 8);
+    asm_x64_write_byte_2(as, REX_PREFIX | REX_W | (src_r64 < 8 ? 0 : REX_R) | (dest_r64 < 8 ? 0 : REX_B), OPCODE_MOV_R64_TO_RM64);
-    asm_x64_write_byte_2(as, REX_PREFIX | REX_W | (src_r64 < 8 ? 0 : REX_R), OPCODE_MOV_R64_TO_RM64);
     asm_x64_write_r64_disp(as, src_r64, dest_r64, dest_disp);
 }
 
@@ -344,8 +349,7 @@ void asm_x64_mov_mem16_to_r64zx(asm_x64_t *as, int src_r64, int src_disp, int de
 
 void asm_x64_mov_mem64_to_r64(asm_x64_t *as, int src_r64, int src_disp, int dest_r64) {
     // use REX prefix for 64 bit operation
-    assert(src_r64 < 8);
+    asm_x64_write_byte_2(as, REX_PREFIX | REX_W | (dest_r64 < 8 ? 0 : REX_R) | (src_r64 < 8 ? 0 : REX_B), OPCODE_MOV_RM64_TO_R64);
-    asm_x64_write_byte_2(as, REX_PREFIX | REX_W | (dest_r64 < 8 ? 0 : REX_R), OPCODE_MOV_RM64_TO_R64);
     asm_x64_write_r64_disp(as, dest_r64, src_r64, src_disp);
 }
 

py/emitnative.c

@@ -140,10 +140,12 @@
 #define ASM_SUB_REG_REG(as, reg_dest, reg_src) asm_x64_sub_r64_r64((as), (reg_dest), (reg_src))
 
 #define ASM_LOAD_REG_REG(as, reg_dest, reg_base) asm_x64_mov_mem64_to_r64((as), (reg_base), 0, (reg_dest))
+#define ASM_LOAD_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_x64_mov_mem64_to_r64((as), (reg_base), 8 * (word_offset), (reg_dest))
 #define ASM_LOAD8_REG_REG(as, reg_dest, reg_base) asm_x64_mov_mem8_to_r64zx((as), (reg_base), 0, (reg_dest))
 #define ASM_LOAD16_REG_REG(as, reg_dest, reg_base) asm_x64_mov_mem16_to_r64zx((as), (reg_base), 0, (reg_dest))
 
 #define ASM_STORE_REG_REG(as, reg_src, reg_base) asm_x64_mov_r64_to_mem64((as), (reg_src), (reg_base), 0)
+#define ASM_STORE_REG_REG_OFFSET(as, reg_src, reg_base, word_offset) asm_x64_mov_r64_to_mem64((as), (reg_src), (reg_base), 8 * (word_offset))
 #define ASM_STORE8_REG_REG(as, reg_src, reg_base) asm_x64_mov_r8_to_mem8((as), (reg_src), (reg_base), 0)
 #define ASM_STORE16_REG_REG(as, reg_src, reg_base) asm_x64_mov_r16_to_mem16((as), (reg_src), (reg_base), 0)
 
@@ -353,10 +355,12 @@ STATIC byte mp_f_n_args[MP_F_NUMBER_OF] = {
 #define ASM_SUB_REG_REG(as, reg_dest, reg_src) asm_thumb_sub_rlo_rlo_rlo((as), (reg_dest), (reg_dest), (reg_src))
 
 #define ASM_LOAD_REG_REG(as, reg_dest, reg_base) asm_thumb_ldr_rlo_rlo_i5((as), (reg_dest), (reg_base), 0)
+#define ASM_LOAD_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_thumb_ldr_rlo_rlo_i5((as), (reg_dest), (reg_base), (word_offset))
 #define ASM_LOAD8_REG_REG(as, reg_dest, reg_base) asm_thumb_ldrb_rlo_rlo_i5((as), (reg_dest), (reg_base), 0)
 #define ASM_LOAD16_REG_REG(as, reg_dest, reg_base) asm_thumb_ldrh_rlo_rlo_i5((as), (reg_dest), (reg_base), 0)
 
 #define ASM_STORE_REG_REG(as, reg_src, reg_base) asm_thumb_str_rlo_rlo_i5((as), (reg_src), (reg_base), 0)
+#define ASM_STORE_REG_REG_OFFSET(as, reg_src, reg_base, word_offset) asm_thumb_str_rlo_rlo_i5((as), (reg_src), (reg_base), (word_offset))
 #define ASM_STORE8_REG_REG(as, reg_src, reg_base) asm_thumb_strb_rlo_rlo_i5((as), (reg_src), (reg_base), 0)
 #define ASM_STORE16_REG_REG(as, reg_src, reg_base) asm_thumb_strh_rlo_rlo_i5((as), (reg_src), (reg_base), 0)
 
@@ -547,6 +551,8 @@ STATIC void emit_native_set_native_type(emit_t *emit, mp_uint_t op, mp_uint_t ar
     }
 }
 
+STATIC void emit_post_push_reg(emit_t *emit, vtype_kind_t vtype, int reg);
+STATIC void emit_native_store_fast(emit_t *emit, qstr qst, mp_uint_t local_num);
 STATIC void emit_native_start_pass(emit_t *emit, pass_kind_t pass, scope_t *scope) {
     DEBUG_printf("start_pass(pass=%u, scope=%p)\n", pass, scope);
 
@@ -669,6 +675,16 @@ STATIC void emit_native_start_pass(emit_t *emit, pass_kind_t pass, scope_t *scop
 #else
     #error not implemented
 #endif
+
+    // initialise closed over variables
+    for (int i = 0; i < scope->id_info_len; i++) {
+        id_info_t *id = &scope->id_info[i];
+        if (id->kind == ID_INFO_KIND_CELL) {
+            ASM_CALL_IND(emit->as, mp_fun_table[MP_F_NEW_CELL], MP_F_NEW_CELL);
+            emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
+            emit_native_store_fast(emit, id->qst, id->local_num);
+        }
+    }
 }
 
 STATIC void emit_native_end_pass(emit_t *emit) {
@@ -1224,12 +1240,15 @@ STATIC void emit_native_load_fast(emit_t *emit, qstr qst, mp_uint_t local_num) {
 }
 
 STATIC void emit_native_load_deref(emit_t *emit, qstr qst, mp_uint_t local_num) {
-    // not implemented
+    DEBUG_printf("load_deref(%s, " UINT_FMT ")\n", qstr_str(qst), local_num);
-    // in principle could support this quite easily (ldr r0, [r0, #0]) and then get closed over variables!
+    need_reg_single(emit, REG_RET, 0);
-    (void)emit;
+    emit_native_load_fast(emit, qst, local_num);
-    (void)qst;
+    vtype_kind_t vtype;
-    (void)local_num;
+    int reg_base = REG_RET;
-    assert(0);
+    emit_pre_pop_reg_flexible(emit, &vtype, &reg_base, -1, -1);
+    ASM_LOAD_REG_REG_OFFSET(emit->as, REG_RET, reg_base, 1);
+    // closed over vars are always Python objects
+    emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
 }
 
 STATIC void emit_native_load_name(emit_t *emit, qstr qst) {
@@ -1446,11 +1465,17 @@ STATIC void emit_native_store_fast(emit_t *emit, qstr qst, mp_uint_t local_num)
 }
 
 STATIC void emit_native_store_deref(emit_t *emit, qstr qst, mp_uint_t local_num) {
-    // not implemented
+    DEBUG_printf("store_deref(%s, " UINT_FMT ")\n", qstr_str(qst), local_num);
-    (void)emit;
+    need_reg_single(emit, REG_TEMP0, 0);
-    (void)qst;
+    need_reg_single(emit, REG_TEMP1, 0);
-    (void)local_num;
+    emit_native_load_fast(emit, qst, local_num);
-    assert(0);
+    vtype_kind_t vtype;
+    int reg_base = REG_TEMP0;
+    emit_pre_pop_reg_flexible(emit, &vtype, &reg_base, -1, -1);
+    int reg_src = REG_TEMP1;
+    emit_pre_pop_reg_flexible(emit, &vtype, &reg_src, reg_base, reg_base);
+    ASM_STORE_REG_REG_OFFSET(emit->as, reg_src, reg_base, 1);
+    emit_post(emit);
 }
 
 STATIC void emit_native_store_name(emit_t *emit, qstr qst) {
@@ -2133,12 +2158,17 @@ STATIC void emit_native_make_function(emit_t *emit, scope_t *scope, mp_uint_t n_
 }
 
 STATIC void emit_native_make_closure(emit_t *emit, scope_t *scope, mp_uint_t n_closed_over, mp_uint_t n_pos_defaults, mp_uint_t n_kw_defaults) {
-    (void)emit;
+    emit_native_pre(emit);
-    (void)scope;
+    if (n_pos_defaults == 0 && n_kw_defaults == 0) {
-    (void)n_closed_over;
+        emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_closed_over);
-    (void)n_pos_defaults;
+        ASM_MOV_IMM_TO_REG(emit->as, n_closed_over, REG_ARG_2);
-    (void)n_kw_defaults;
+    } else {
-    assert(0);
+        emit_get_stack_pointer_to_reg_for_pop(emit, REG_ARG_3, n_closed_over + 2);
+        ASM_MOV_IMM_TO_REG(emit->as, 0x100 | n_closed_over, REG_ARG_2);
+    }
+    ASM_MOV_ALIGNED_IMM_TO_REG(emit->as, (mp_uint_t)scope->raw_code, REG_ARG_1);
+    ASM_CALL_IND(emit->as, mp_fun_table[MP_F_MAKE_CLOSURE_FROM_RAW_CODE], MP_F_MAKE_CLOSURE_FROM_RAW_CODE);
+    emit_post_push_reg(emit, VTYPE_PYOBJ, REG_RET);
 }
 
 STATIC void emit_native_call_function(emit_t *emit, mp_uint_t n_positional, mp_uint_t n_keyword, mp_uint_t star_flags) {

py/nativeglue.c

@@ -132,6 +132,8 @@ void *const mp_fun_table[MP_F_NUMBER_OF] = {
     mp_unpack_ex,
     mp_delete_name,
     mp_delete_global,
+    mp_obj_new_cell,
+    mp_make_closure_from_raw_code,
 };
 
 /*

py/runtime0.h

@@ -148,6 +148,8 @@ typedef enum {
     MP_F_UNPACK_EX,
     MP_F_DELETE_NAME,
     MP_F_DELETE_GLOBAL,
+    MP_F_NEW_CELL,
+    MP_F_MAKE_CLOSURE_FROM_RAW_CODE,
     MP_F_NUMBER_OF,
 } mp_fun_kind_t;
 

tests/run-tests

@@ -153,7 +153,7 @@ def run_tests(pyb, tests, args):
     # Some tests are known to fail with native emitter
     # Remove them from the below when they work
     if args.emit == 'native':
-        skip_tests.update({'basics/%s.py' % t for t in 'bytes_gen class_store_class class_super class_super_object closure1 closure2 closure_defargs del_deref del_local fun3 fun_calldblstar fun_callstar fun_callstardblstar fun_defargs fun_defargs2 fun_kwargs fun_kwonly fun_kwonlydef fun_kwvarargs fun_varargs gen_yield_from gen_yield_from_close gen_yield_from_ducktype gen_yield_from_exc gen_yield_from_iter gen_yield_from_send gen_yield_from_throw generator1 generator2 generator_args generator_close generator_closure generator_exc generator_return generator_send globals_del string_format string_join subclass_native2_list subclass_native2_tuple try_finally_loops try_finally_return try_reraise try_reraise2 unboundlocal with1 with_break with_continue with_return'.split()})
+        skip_tests.update({'basics/%s.py' % t for t in 'bytes_gen class_store_class closure_defargs del_deref del_local fun3 fun_calldblstar fun_callstar fun_callstardblstar fun_defargs fun_defargs2 fun_kwargs fun_kwonly fun_kwonlydef fun_kwvarargs fun_varargs gen_yield_from gen_yield_from_close gen_yield_from_ducktype gen_yield_from_exc gen_yield_from_iter gen_yield_from_send gen_yield_from_throw generator1 generator2 generator_args generator_close generator_closure generator_exc generator_return generator_send globals_del string_format string_join subclass_native2_list subclass_native2_tuple try_finally_loops try_finally_return try_reraise try_reraise2 unboundlocal with1 with_break with_continue with_return'.split()})
         skip_tests.add('basics/array_construct2.py') # requires generators
         skip_tests.add('basics/bool1.py') # seems to randomly fail
         skip_tests.add('basics/boundmeth1.py') # requires support for many args