py/asmxtensa: Fix use of l32i/s32i when offset won't fit in encoding.

This commit adds optimised l32i/s32i functions that select the best load/
store encoding based on the size of the offset, and uses the function when
necessary in code generation.

Without this, ASM_LOAD_REG_REG_OFFSET() could overflow the word offset
(using a narrow encoding), for example when loading the prelude from the
constant table when there are many (>16) constants.

Fixes issue #8458.

Signed-off-by: Damien George <damien@micropython.org>

2 files changed, 26 insertions, 12 deletions

diff --git a/py/asmxtensa.c b/py/asmxtensa.c
index 0956d50f3..8ac914ec4 100644
--- a/py/asmxtensa.c
+++ b/py/asmxtensa.c
@@ -27,7 +27,7 @@
 #include <stdio.h>
 #include <assert.h>
 
-#include "py/mpconfig.h"
+#include "py/runtime.h"
 
 // wrapper around everything in this file
 #if MICROPY_EMIT_XTENSA || MICROPY_EMIT_INLINE_XTENSA || MICROPY_EMIT_XTENSAWIN
@@ -232,21 +232,33 @@ void asm_xtensa_mov_reg_pcrel(asm_xtensa_t *as, uint reg_dest, uint label) {
     asm_xtensa_op_add_n(as, reg_dest, reg_dest, ASM_XTENSA_REG_A0);
 }
 
-void asm_xtensa_call_ind(asm_xtensa_t *as, uint idx) {
-    if (idx < 16) {
-        asm_xtensa_op_l32i_n(as, ASM_XTENSA_REG_A0, ASM_XTENSA_REG_FUN_TABLE, idx);
+void asm_xtensa_l32i_optimised(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint word_offset) {
+    if (word_offset < 16) {
+        asm_xtensa_op_l32i_n(as, reg_dest, reg_base, word_offset);
+    } else if (word_offset < 256) {
+        asm_xtensa_op_l32i(as, reg_dest, reg_base, word_offset);
     } else {
-        asm_xtensa_op_l32i(as, ASM_XTENSA_REG_A0, ASM_XTENSA_REG_FUN_TABLE, idx);
+        mp_raise_msg(&mp_type_RuntimeError, MP_ERROR_TEXT("asm overflow"));
     }
-    asm_xtensa_op_callx0(as, ASM_XTENSA_REG_A0);
 }
 
-void asm_xtensa_call_ind_win(asm_xtensa_t *as, uint idx) {
-    if (idx < 16) {
-        asm_xtensa_op_l32i_n(as, ASM_XTENSA_REG_A8, ASM_XTENSA_REG_FUN_TABLE_WIN, idx);
+void asm_xtensa_s32i_optimised(asm_xtensa_t *as, uint reg_src, uint reg_base, uint word_offset) {
+    if (word_offset < 16) {
+        asm_xtensa_op_s32i_n(as, reg_src, reg_base, word_offset);
+    } else if (word_offset < 256) {
+        asm_xtensa_op_s32i(as, reg_src, reg_base, word_offset);
     } else {
-        asm_xtensa_op_l32i(as, ASM_XTENSA_REG_A8, ASM_XTENSA_REG_FUN_TABLE_WIN, idx);
+        mp_raise_msg(&mp_type_RuntimeError, MP_ERROR_TEXT("asm overflow"));
     }
+}
+
+void asm_xtensa_call_ind(asm_xtensa_t *as, uint idx) {
+    asm_xtensa_l32i_optimised(as, ASM_XTENSA_REG_A0, ASM_XTENSA_REG_FUN_TABLE, idx);
+    asm_xtensa_op_callx0(as, ASM_XTENSA_REG_A0);
+}
+
+void asm_xtensa_call_ind_win(asm_xtensa_t *as, uint idx) {
+    asm_xtensa_l32i_optimised(as, ASM_XTENSA_REG_A8, ASM_XTENSA_REG_FUN_TABLE_WIN, idx);
     asm_xtensa_op_callx8(as, ASM_XTENSA_REG_A8);
 }
 
diff --git a/py/asmxtensa.h b/py/asmxtensa.h
index 43f1b608e..16a59c0ee 100644
--- a/py/asmxtensa.h
+++ b/py/asmxtensa.h
@@ -278,6 +278,8 @@ void asm_xtensa_mov_local_reg(asm_xtensa_t *as, int local_num, uint reg_src);
 void asm_xtensa_mov_reg_local(asm_xtensa_t *as, uint reg_dest, int local_num);
 void asm_xtensa_mov_reg_local_addr(asm_xtensa_t *as, uint reg_dest, int local_num);
 void asm_xtensa_mov_reg_pcrel(asm_xtensa_t *as, uint reg_dest, uint label);
+void asm_xtensa_l32i_optimised(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint word_offset);
+void asm_xtensa_s32i_optimised(asm_xtensa_t *as, uint reg_src, uint reg_base, uint word_offset);
 void asm_xtensa_call_ind(asm_xtensa_t *as, uint idx);
 void asm_xtensa_call_ind_win(asm_xtensa_t *as, uint idx);
 
@@ -393,12 +395,12 @@ void asm_xtensa_call_ind_win(asm_xtensa_t *as, uint idx);
 #define ASM_SUB_REG_REG(as, reg_dest, reg_src) asm_xtensa_op_sub((as), (reg_dest), (reg_dest), (reg_src))
 #define ASM_MUL_REG_REG(as, reg_dest, reg_src) asm_xtensa_op_mull((as), (reg_dest), (reg_dest), (reg_src))
 
-#define ASM_LOAD_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_op_l32i_n((as), (reg_dest), (reg_base), (word_offset))
+#define ASM_LOAD_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_l32i_optimised((as), (reg_dest), (reg_base), (word_offset))
 #define ASM_LOAD8_REG_REG(as, reg_dest, reg_base) asm_xtensa_op_l8ui((as), (reg_dest), (reg_base), 0)
 #define ASM_LOAD16_REG_REG(as, reg_dest, reg_base) asm_xtensa_op_l16ui((as), (reg_dest), (reg_base), 0)
 #define ASM_LOAD32_REG_REG(as, reg_dest, reg_base) asm_xtensa_op_l32i_n((as), (reg_dest), (reg_base), 0)
 
-#define ASM_STORE_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_op_s32i_n((as), (reg_dest), (reg_base), (word_offset))
+#define ASM_STORE_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_s32i_optimised((as), (reg_dest), (reg_base), (word_offset))
 #define ASM_STORE8_REG_REG(as, reg_src, reg_base) asm_xtensa_op_s8i((as), (reg_src), (reg_base), 0)
 #define ASM_STORE16_REG_REG(as, reg_src, reg_base) asm_xtensa_op_s16i((as), (reg_src), (reg_base), 0)
 #define ASM_STORE32_REG_REG(as, reg_src, reg_base) asm_xtensa_op_s32i_n((as), (reg_src), (reg_base), 0)