py/emitnative: Access qstr values using indirection table qstr_table.

dpgeorge · dpgeorge · commit d4d53e9e114d · 2022-05-23T15:43:06.000+10:00
This changes the native emitter to access qstr values using the qstr
indirection table qstr_table, but only when generating native code that
will be saved to a .mpy file.  This makes the resulting native code fully
static, ie it does not require any fix-ups or rewriting when it is
imported.

The performance of native code is more or less unchanged.  Benchmark
results on PYBv1.0 (using --via-mpy and --emit native) are:

N=100 M=100          baseline -&gt; this-commit     diff      diff% (error%)
bm_chaos.py            407.16 -&gt;     411.85 :   +4.69 =  +1.152% (+/-0.01%)
bm_fannkuch.py         100.89 -&gt;     101.20 :   +0.31 =  +0.307% (+/-0.01%)
bm_fft.py             3521.17 -&gt;    3441.72 :  -79.45 =  -2.256% (+/-0.00%)
bm_float.py           6707.29 -&gt;    6644.83 :  -62.46 =  -0.931% (+/-0.00%)
bm_hexiom.py            55.91 -&gt;      55.41 :   -0.50 =  -0.894% (+/-0.00%)
bm_nqueens.py         5343.54 -&gt;    5326.17 :  -17.37 =  -0.325% (+/-0.00%)
bm_pidigits.py         603.89 -&gt;     632.79 :  +28.90 =  +4.786% (+/-0.33%)
core_qstr.py            64.18 -&gt;      64.09 :   -0.09 =  -0.140% (+/-0.01%)
core_yield_from.py     313.61 -&gt;     311.11 :   -2.50 =  -0.797% (+/-0.03%)
misc_aes.py            654.29 -&gt;     659.75 :   +5.46 =  +0.834% (+/-0.02%)
misc_mandel.py        4205.10 -&gt;    4272.08 :  +66.98 =  +1.593% (+/-0.01%)
misc_pystone.py       3077.79 -&gt;    3128.39 :  +50.60 =  +1.644% (+/-0.01%)
misc_raytrace.py       388.45 -&gt;     393.71 :   +5.26 =  +1.354% (+/-0.01%)
viper_call0.py         576.83 -&gt;     566.76 :  -10.07 =  -1.746% (+/-0.05%)
viper_call1a.py        550.39 -&gt;     540.12 :  -10.27 =  -1.866% (+/-0.11%)
viper_call1b.py        438.32 -&gt;     432.09 :   -6.23 =  -1.421% (+/-0.11%)
viper_call1c.py        442.96 -&gt;     436.11 :   -6.85 =  -1.546% (+/-0.08%)
viper_call2a.py        536.31 -&gt;     527.37 :   -8.94 =  -1.667% (+/-0.04%)
viper_call2b.py        378.99 -&gt;     377.50 :   -1.49 =  -0.393% (+/-0.08%)

Signed-off-by: Damien George &lt;damien@micropython.org&gt;
diff --git a/py/compile.c b/py/compile.c
@@ -3323,7 +3323,7 @@ STATIC void compile_scope_inline_asm(compiler_t *comp, scope_t *scope, pass_kind
                 NULL,
                 #if MICROPY_PERSISTENT_CODE_SAVE
                 0,
-                0, 0, NULL,
+                0,
                 #endif
                 0, comp->scope_cur->num_pos_args, type_sig);
         }
diff --git a/py/emitglue.c b/py/emitglue.c
@@ -101,7 +101,6 @@ void mp_emit_glue_assign_native(mp_raw_code_t *rc, mp_raw_code_kind_t kind, void
     #if MICROPY_PERSISTENT_CODE_SAVE
     size_t n_children,
     uint16_t prelude_offset,
-    uint16_t n_qstr, mp_qstr_link_entry_t *qstr_link,
     #endif
     mp_uint_t scope_flags, mp_uint_t n_pos_args, mp_uint_t type_sig) {
 
@@ -144,8 +143,6 @@ void mp_emit_glue_assign_native(mp_raw_code_t *rc, mp_raw_code_kind_t kind, void
     #if MICROPY_PERSISTENT_CODE_SAVE
     rc->n_children = n_children;
     rc->prelude_offset = prelude_offset;
-    rc->n_qstr = n_qstr;
-    rc->qstr_link = qstr_link;
     #endif
 
     // These two entries are only needed for MP_CODE_NATIVE_ASM.
diff --git a/py/emitglue.h b/py/emitglue.h
@@ -49,11 +49,6 @@ typedef enum {
     MP_CODE_NATIVE_ASM,
 } mp_raw_code_kind_t;
 
-typedef struct _mp_qstr_link_entry_t {
-    uint16_t off;
-    uint16_t qst;
-} mp_qstr_link_entry_t;
-
 // compiled bytecode: instance in RAM, referenced by outer scope, usually freed after first (and only) use
 // mpy file: instance in RAM, created when .mpy file is loaded (same comments as above)
 // frozen: instance in ROM
@@ -78,8 +73,6 @@ typedef struct _mp_raw_code_t {
     #endif
     #if MICROPY_EMIT_MACHINE_CODE
     uint16_t prelude_offset;
-    uint16_t n_qstr;
-    mp_qstr_link_entry_t *qstr_link;
     #endif
     #endif
     #if MICROPY_EMIT_MACHINE_CODE
@@ -104,7 +97,6 @@ void mp_emit_glue_assign_native(mp_raw_code_t *rc, mp_raw_code_kind_t kind, void
     #if MICROPY_PERSISTENT_CODE_SAVE
     size_t n_children,
     uint16_t prelude_offset,
-    uint16_t n_qstr, mp_qstr_link_entry_t *qstr_link,
     #endif
     mp_uint_t scope_flags, mp_uint_t n_pos_args, mp_uint_t type_sig);
 
diff --git a/py/emitnarm.c b/py/emitnarm.c
@@ -10,8 +10,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (3) // r4
-#define NLR_BUF_IDX_LOCAL_2 (4) // r5
-#define NLR_BUF_IDX_LOCAL_3 (5) // r6
 
 #define N_ARM (1)
 #define EXPORT_FUN(name) emit_native_arm_##name
diff --git a/py/emitnative.c b/py/emitnative.c
diff --git a/py/emitnthumb.c b/py/emitnthumb.c
@@ -10,8 +10,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (3) // r4
-#define NLR_BUF_IDX_LOCAL_2 (4) // r5
-#define NLR_BUF_IDX_LOCAL_3 (5) // r6
 
 #define N_THUMB (1)
 #define EXPORT_FUN(name) emit_native_thumb_##name
diff --git a/py/emitnx64.c b/py/emitnx64.c
@@ -10,8 +10,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (5) // rbx
-#define NLR_BUF_IDX_LOCAL_2 (6) // r12
-#define NLR_BUF_IDX_LOCAL_3 (7) // r13
 
 #define N_X64 (1)
 #define EXPORT_FUN(name) emit_native_x64_##name
diff --git a/py/emitnx86.c b/py/emitnx86.c
@@ -11,8 +11,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (5) // ebx
-#define NLR_BUF_IDX_LOCAL_2 (7) // esi
-#define NLR_BUF_IDX_LOCAL_3 (6) // edi
 
 // x86 needs a table to know how many args a given function has
 STATIC byte mp_f_n_args[MP_F_NUMBER_OF] = {
diff --git a/py/emitnxtensa.c b/py/emitnxtensa.c
@@ -10,8 +10,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (8) // a12
-#define NLR_BUF_IDX_LOCAL_2 (9) // a13
-#define NLR_BUF_IDX_LOCAL_3 (10) // a14
 
 #define N_XTENSA (1)
 #define EXPORT_FUN(name) emit_native_xtensa_##name
diff --git a/py/emitnxtensawin.c b/py/emitnxtensawin.c
@@ -11,8 +11,6 @@
 
 // Word indices of REG_LOCAL_x in nlr_buf_t
 #define NLR_BUF_IDX_LOCAL_1 (2 + 4) // a4
-#define NLR_BUF_IDX_LOCAL_2 (2 + 5) // a5
-#define NLR_BUF_IDX_LOCAL_3 (2 + 6) // a6
 
 #define N_NLR_SETJMP (1)
 #define N_XTENSAWIN (1)
diff --git a/py/persistentcode.c b/py/persistentcode.c
@@ -75,38 +75,6 @@ typedef struct _reloc_info_t {
     uint8_t *bss;
 } reloc_info_t;
 
-#if MICROPY_EMIT_THUMB
-STATIC void asm_thumb_rewrite_mov(uint8_t *pc, uint16_t val) {
-    // high part
-    *(uint16_t *)pc = (*(uint16_t *)pc & 0xfbf0) | (val >> 1 & 0x0400) | (val >> 12);
-    // low part
-    *(uint16_t *)(pc + 2) = (*(uint16_t *)(pc + 2) & 0x0f00) | (val << 4 & 0x7000) | (val & 0x00ff);
-
-}
-#endif
-
-STATIC void arch_link_qstr(uint8_t *pc, bool is_obj, qstr qst) {
-    mp_uint_t val = qst;
-    if (is_obj) {
-        val = (mp_uint_t)MP_OBJ_NEW_QSTR(qst);
-    }
-    #if MICROPY_EMIT_X86 || MICROPY_EMIT_X64 || MICROPY_EMIT_ARM || MICROPY_EMIT_XTENSA || MICROPY_EMIT_XTENSAWIN
-    pc[0] = val & 0xff;
-    pc[1] = (val >> 8) & 0xff;
-    pc[2] = (val >> 16) & 0xff;
-    pc[3] = (val >> 24) & 0xff;
-    #elif MICROPY_EMIT_THUMB
-    if (is_obj) {
-        // qstr object, movw and movt
-        asm_thumb_rewrite_mov(pc, val); // movw
-        asm_thumb_rewrite_mov(pc + 4, val >> 16); // movt
-    } else {
-        // qstr number, movw instruction
-        asm_thumb_rewrite_mov(pc, val); // movw
-    }
-    #endif
-}
-
 void mp_native_relocate(void *ri_in, uint8_t *text, uintptr_t reloc_text) {
     // Relocate native code
     reloc_info_t *ri = ri_in;
@@ -285,9 +253,6 @@ STATIC mp_raw_code_t *load_raw_code(mp_reader_t *reader) {
                 } else if ((off & 3) == 3) {
                     // Generic, aligned qstr-object link
                     *(mp_obj_t *)dest = MP_OBJ_NEW_QSTR(qst);
-                } else {
-                    // Architecture-specific link
-                    arch_link_qstr(dest, (off & 3) == 2, qst);
                 }
             }
         }
@@ -424,7 +389,6 @@ STATIC mp_raw_code_t *load_raw_code(mp_reader_t *reader) {
             #if MICROPY_PERSISTENT_CODE_SAVE
             n_children,
             prelude_offset,
-            0, NULL,
             #endif
             native_scope_flags, native_n_pos_args, native_type_sig
             );
@@ -605,11 +569,7 @@ STATIC void save_raw_code(mp_print_t *print, const mp_raw_code_t *rc) {
     #if MICROPY_EMIT_MACHINE_CODE
     if (rc->kind == MP_CODE_NATIVE_PY || rc->kind == MP_CODE_NATIVE_VIPER) {
         // Save qstr link table for native code
-        mp_print_uint(print, rc->n_qstr);
-        for (size_t i = 0; i < rc->n_qstr; ++i) {
-            mp_print_uint(print, rc->qstr_link[i].off);
-            save_qstr(print, rc->qstr_link[i].qst);
-        }
+        mp_print_uint(print, 0);
     }
 
     if (rc->kind == MP_CODE_NATIVE_PY) {
diff --git a/tools/mpy-tool.py b/tools/mpy-tool.py
@@ -839,7 +839,7 @@ def freeze_children(self, prelude_ptr=None):
             print("};")
             print()
 
-    def freeze_raw_code(self, prelude_ptr=None, qstr_links=(), type_sig=0):
+    def freeze_raw_code(self, prelude_ptr=None, type_sig=0):
         # Generate mp_raw_code_t.
         print("static const mp_raw_code_t raw_code_%s = {" % self.escaped_name)
         print("    .kind = %s," % RawCode.code_kind_str[self.code_kind])
@@ -879,8 +879,6 @@ def freeze_raw_code(self, prelude_ptr=None, qstr_links=(), type_sig=0):
             print("    #endif")
         print("    #if MICROPY_EMIT_MACHINE_CODE")
         print("    .prelude_offset = %u," % self.prelude_offset)
-        print("    .n_qstr = %u," % len(qstr_links))
-        print("    .qstr_link = NULL,")  # TODO
         print("    #endif")
         print("    #endif")
         print("    #if MICROPY_EMIT_MACHINE_CODE")
@@ -1038,47 +1036,6 @@ def disassemble(self):
             ip += sz
         self.disassemble_children()
 
-    def _asm_thumb_rewrite_mov(self, pc, val):
-        print("    (%u & 0xf0) | (%s >> 12)," % (self.fun_data[pc], val), end="")
-        print(" (%u & 0xfb) | (%s >> 9 & 0x04)," % (self.fun_data[pc + 1], val), end="")
-        print(" (%s & 0xff)," % (val,), end="")
-        print(" (%u & 0x07) | (%s >> 4 & 0x70)," % (self.fun_data[pc + 3], val))
-
-    def _link_qstr(self, pc, kind, qst):
-        if kind == 0:
-            # Generic 16-bit link
-            print("    %s & 0xff, %s >> 8," % (qst, qst))
-            return 2
-        else:
-            # Architecture-specific link
-            is_obj = kind == 2
-            if is_obj:
-                qst = "((uintptr_t)MP_OBJ_NEW_QSTR(%s))" % qst
-            if config.native_arch in (
-                MP_NATIVE_ARCH_X86,
-                MP_NATIVE_ARCH_X64,
-                MP_NATIVE_ARCH_ARMV6,
-                MP_NATIVE_ARCH_XTENSA,
-                MP_NATIVE_ARCH_XTENSAWIN,
-            ):
-                print(
-                    "    %s & 0xff, (%s >> 8) & 0xff, (%s >> 16) & 0xff, %s >> 24,"
-                    % (qst, qst, qst, qst)
-                )
-                return 4
-            elif MP_NATIVE_ARCH_ARMV6M <= config.native_arch <= MP_NATIVE_ARCH_ARMV7EMDP:
-                if is_obj:
-                    # qstr object, movw and movt
-                    self._asm_thumb_rewrite_mov(pc, qst)
-                    self._asm_thumb_rewrite_mov(pc + 4, "(%s >> 16)" % qst)
-                    return 8
-                else:
-                    # qstr number, movw instruction
-                    self._asm_thumb_rewrite_mov(pc, qst)
-                    return 4
-            else:
-                assert 0
-
     def freeze(self):
         if self.scope_flags & ~0x0F:
             raise FreezeError("unable to freeze code with relocations")
@@ -1098,21 +1055,13 @@ def freeze(self):
         i = 0
         qi = 0
         while i < i_top:
-            if qi < len(self.qstr_links) and i == self.qstr_links[qi][0]:
-                # link qstr
-                qi_off, qi_kind, qi_val = self.qstr_links[qi]
-                i += self._link_qstr(i, qi_kind, qi_val.qstr_id)
-                qi += 1
-            else:
-                # copy machine code (max 16 bytes)
-                i16 = min(i + 16, i_top)
-                if qi < len(self.qstr_links):
-                    i16 = min(i16, self.qstr_links[qi][0])
-                print("   ", end="")
-                for ii in range(i, i16):
-                    print(" 0x%02x," % self.fun_data[ii], end="")
-                print()
-                i = i16
+            # copy machine code (max 16 bytes)
+            i16 = min(i + 16, i_top)
+            print("   ", end="")
+            for ii in range(i, i16):
+                print(" 0x%02x," % self.fun_data[ii], end="")
+            print()
+            i = i16
 
         print("};")
 
@@ -1134,7 +1083,7 @@ def freeze(self):
             print("#endif")
 
         self.freeze_children(prelude_ptr)
-        self.freeze_raw_code(prelude_ptr, self.qstr_links, self.type_sig)
+        self.freeze_raw_code(prelude_ptr, self.type_sig)
 
 
 class MPYSegment:

Original file line number	Diff line number	Diff line change
`@@ -3323,7 +3323,7 @@ STATIC void compile_scope_inline_asm(compiler_t comp, scope_t scope, pass_kind`
`3323`	`3323`	`NULL,`
`3324`	`3324`	`#if MICROPY_PERSISTENT_CODE_SAVE`
`3325`	`3325`	`0,`
`3326`		`- 0, 0, NULL,`
	`3326`	`+ 0,`
`3327`	`3327`	`#endif`
`3328`	`3328`	`0, comp->scope_cur->num_pos_args, type_sig);`
`3329`	`3329`	`}`