py/asmxtensa: Implement the full set of Viper load/store operations.

This commit expands the implementation of Viper load/store operations
that are optimised for the Xtensa platform.

Now both load and store emitters should generate the shortest possible
sequence in all cases.  Redundant specialised operation emitters have
been aliased to the general case implementation - this was the case of
integer-indexed load/store operations with a fixed offset of zero.

Signed-off-by: Alessandro Gatti <a.gatti@frob.it>

3 files changed, 60 insertions, 44 deletions

diff --git a/py/asmxtensa.c b/py/asmxtensa.c
index 85a8cfef5..bc3e717d9 100644
--- a/py/asmxtensa.c
+++ b/py/asmxtensa.c
@@ -299,25 +299,47 @@ void asm_xtensa_l32i_optimised(asm_xtensa_t *as, uint reg_dest, uint reg_base, u
     }
 }
 
-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);
+void asm_xtensa_load_reg_reg_offset(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint offset, uint operation_size) {
+    assert(operation_size <= 2 && "Operation size value out of range.");
+
+    if (operation_size == 2 && MP_FIT_UNSIGNED(4, offset)) {
+        asm_xtensa_op_l32i_n(as, reg_dest, reg_base, offset);
+        return;
+    }
+
+    if (MP_FIT_UNSIGNED(8, offset)) {
+        asm_xtensa_op24(as, ASM_XTENSA_ENCODE_RRI8(2, operation_size, reg_base, reg_dest, offset));
+        return;
+    }
+
+    asm_xtensa_mov_reg_i32_optimised(as, reg_dest, offset << operation_size);
+    asm_xtensa_op_add_n(as, reg_dest, reg_base, reg_dest);
+    if (operation_size == 2) {
+        asm_xtensa_op_l32i_n(as, reg_dest, reg_dest, 0);
     } else {
-        asm_xtensa_mov_reg_i32_optimised(as, REG_TEMP, word_offset * 4);
-        asm_xtensa_op_add_n(as, REG_TEMP, reg_base, REG_TEMP);
-        asm_xtensa_op_s32i_n(as, reg_src, REG_TEMP, 0);
+        asm_xtensa_op24(as, ASM_XTENSA_ENCODE_RRI8(2, operation_size, reg_dest, reg_dest, 0));
     }
 }
 
-void asm_xtensa_l16ui_optimised(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint halfword_offset) {
-    if (halfword_offset < 256) {
-        asm_xtensa_op_l16ui(as, reg_dest, reg_base, halfword_offset);
+void asm_xtensa_store_reg_reg_offset(asm_xtensa_t *as, uint reg_src, uint reg_base, uint offset, uint operation_size) {
+    assert(operation_size <= 2 && "Operation size value out of range.");
+
+    if (operation_size == 2 && MP_FIT_UNSIGNED(4, offset)) {
+        asm_xtensa_op_s32i_n(as, reg_src, reg_base, offset);
+        return;
+    }
+
+    if (MP_FIT_UNSIGNED(8, offset)) {
+        asm_xtensa_op24(as, ASM_XTENSA_ENCODE_RRI8(2, 0x04 | operation_size, reg_base, reg_src, offset));
+        return;
+    }
+
+    asm_xtensa_mov_reg_i32_optimised(as, REG_TEMP, offset << operation_size);
+    asm_xtensa_op_add_n(as, REG_TEMP, reg_base, REG_TEMP);
+    if (operation_size == 2) {
+        asm_xtensa_op_s32i_n(as, reg_src, REG_TEMP, 0);
     } else {
-        asm_xtensa_mov_reg_i32_optimised(as, reg_dest, halfword_offset * 2);
-        asm_xtensa_op_add_n(as, reg_dest, reg_base, reg_dest);
-        asm_xtensa_op_l16ui(as, reg_dest, reg_dest, 0);
+        asm_xtensa_op24(as, ASM_XTENSA_ENCODE_RRI8(2, 0x04 | operation_size, REG_TEMP, reg_src, 0));
     }
 }
 
diff --git a/py/asmxtensa.h b/py/asmxtensa.h
index d90aef3c5..7f113ca12 100644
--- a/py/asmxtensa.h
+++ b/py/asmxtensa.h
@@ -293,9 +293,8 @@ 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_l16ui_optimised(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint halfword_offset);
+void asm_xtensa_load_reg_reg_offset(asm_xtensa_t *as, uint reg_dest, uint reg_base, uint offset, uint operation_size);
+void asm_xtensa_store_reg_reg_offset(asm_xtensa_t *as, uint reg_src, uint reg_base, uint offset, uint operation_size);
 void asm_xtensa_call_ind(asm_xtensa_t *as, uint idx);
 void asm_xtensa_call_ind_win(asm_xtensa_t *as, uint idx);
 void asm_xtensa_bit_branch(asm_xtensa_t *as, mp_uint_t reg, mp_uint_t bit, mp_uint_t label, mp_uint_t condition);
@@ -420,16 +419,22 @@ void asm_xtensa_l32r(asm_xtensa_t *as, mp_uint_t reg, mp_uint_t label);
 #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_LOAD32_REG_REG_OFFSET((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_LOAD16_REG_REG_OFFSET(as, reg_dest, reg_base, uint16_offset) asm_xtensa_l16ui_optimised((as), (reg_dest), (reg_base), (uint16_offset))
+#define ASM_LOAD8_REG_REG(as, reg_dest, reg_base) ASM_LOAD8_REG_REG_OFFSET((as), (reg_dest), (reg_base), 0)
+#define ASM_LOAD8_REG_REG_OFFSET(as, reg_dest, reg_base, byte_offset) asm_xtensa_load_reg_reg_offset((as), (reg_dest), (reg_base), (byte_offset), 0)
+#define ASM_LOAD8_REG_REG_REG(as, reg_dest, reg_base, reg_index) \
+    do { \
+        asm_xtensa_op_add_n((as), (reg_base), (reg_index), (reg_base)); \
+        asm_xtensa_op_l8ui((as), (reg_dest), (reg_base), 0); \
+    } while (0)
+#define ASM_LOAD16_REG_REG(as, reg_dest, reg_base) ASM_LOAD16_REG_REG_OFFSET((as), (reg_dest), (reg_base), 0)
+#define ASM_LOAD16_REG_REG_OFFSET(as, reg_dest, reg_base, halfword_offset) asm_xtensa_load_reg_reg_offset((as), (reg_dest), (reg_base), (halfword_offset), 1)
 #define ASM_LOAD16_REG_REG_REG(as, reg_dest, reg_base, reg_index) \
     do { \
         asm_xtensa_op_addx2((as), (reg_base), (reg_index), (reg_base)); \
         asm_xtensa_op_l16ui((as), (reg_dest), (reg_base), 0); \
     } while (0)
-#define ASM_LOAD32_REG_REG(as, reg_dest, reg_base) asm_xtensa_op_l32i_n((as), (reg_dest), (reg_base), 0)
-#define ASM_LOAD32_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_l32i_optimised((as), (reg_dest), (reg_base), (word_offset))
+#define ASM_LOAD32_REG_REG(as, reg_dest, reg_base) ASM_LOAD32_REG_REG_OFFSET((as), (reg_dest), (reg_base), 0)
+#define ASM_LOAD32_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_load_reg_reg_offset((as), (reg_dest), (reg_base), (word_offset), 2)
 #define ASM_LOAD32_REG_REG_REG(as, reg_dest, reg_base, reg_index) \
     do { \
         asm_xtensa_op_addx4((as), (reg_base), (reg_index), (reg_base)); \
@@ -437,15 +442,22 @@ void asm_xtensa_l32r(asm_xtensa_t *as, mp_uint_t reg, mp_uint_t label);
     } while (0)
 
 #define ASM_STORE_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) ASM_STORE32_REG_REG_OFFSET((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_STORE8_REG_REG(as, reg_src, reg_base) ASM_STORE8_REG_REG_OFFSET((as), (reg_src), (reg_base), 0)
+#define ASM_STORE8_REG_REG_OFFSET(as, reg_src, reg_base, byte_offset) asm_xtensa_store_reg_reg_offset((as), (reg_src), (reg_base), (byte_offset), 0)
+#define ASM_STORE8_REG_REG_REG(as, reg_val, reg_base, reg_index) \
+    do { \
+        asm_xtensa_op_add_n((as), (reg_base), (reg_index), (reg_base)); \
+        asm_xtensa_op_s8i((as), (reg_val), (reg_base), 0); \
+    } while (0)
+#define ASM_STORE16_REG_REG(as, reg_src, reg_base) ASM_STORE16_REG_REG_OFFSET((as), (reg_src), (reg_base), 0)
+#define ASM_STORE16_REG_REG_OFFSET(as, reg_src, reg_base, halfword_offset) asm_xtensa_store_reg_reg_offset((as), (reg_src), (reg_base), (halfword_offset), 1)
 #define ASM_STORE16_REG_REG_REG(as, reg_val, reg_base, reg_index) \
     do { \
         asm_xtensa_op_addx2((as), (reg_base), (reg_index), (reg_base)); \
         asm_xtensa_op_s16i((as), (reg_val), (reg_base), 0); \
     } while (0)
-#define ASM_STORE32_REG_REG(as, reg_src, reg_base) asm_xtensa_op_s32i_n((as), (reg_src), (reg_base), 0)
-#define ASM_STORE32_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_s32i_optimised((as), (reg_dest), (reg_base), (word_offset))
+#define ASM_STORE32_REG_REG(as, reg_src, reg_base) ASM_STORE32_REG_REG_OFFSET((as), (reg_src), (reg_base), 0)
+#define ASM_STORE32_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_xtensa_store_reg_reg_offset((as), (reg_dest), (reg_base), (word_offset), 2)
 #define ASM_STORE32_REG_REG_REG(as, reg_val, reg_base, reg_index) \
     do { \
         asm_xtensa_op_addx4((as), (reg_base), (reg_index), (reg_base)); \
diff --git a/py/emitnative.c b/py/emitnative.c
index 36e9719db..f3ab483e8 100644
--- a/py/emitnative.c
+++ b/py/emitnative.c
@@ -1534,12 +1534,6 @@ static void emit_native_load_subscr(emit_t *emit) {
                     #ifdef ASM_LOAD8_REG_REG_OFFSET
                     ASM_LOAD8_REG_REG_OFFSET(emit->as, REG_RET, reg_base, index_value);
                     #else
-                    #if N_XTENSA || N_XTENSAWIN
-                    if (index_value >= 0 && index_value < 256) {
-                        asm_xtensa_op_l8ui(emit->as, REG_RET, reg_base, index_value);
-                        break;
-                    }
-                    #endif
                     if (index_value != 0) {
                         // index is non-zero
                         need_reg_single(emit, reg_index, 0);
@@ -1776,12 +1770,6 @@ static void emit_native_store_subscr(emit_t *emit) {
                     #ifdef ASM_STORE8_REG_REG_OFFSET
                     ASM_STORE8_REG_REG_OFFSET(emit->as, reg_value, reg_base, index_value);
                     #else
-                    #if N_XTENSA || N_XTENSAWIN
-                    if (index_value >= 0 && index_value < 256) {
-                        asm_xtensa_op_s8i(emit->as, reg_value, reg_base, index_value);
-                        break;
-                    }
-                    #endif
                     if (index_value != 0) {
                         // index is non-zero
                         ASM_MOV_REG_IMM(emit->as, reg_index, index_value);
@@ -1797,12 +1785,6 @@ static void emit_native_store_subscr(emit_t *emit) {
                     #ifdef ASM_STORE16_REG_REG_OFFSET
                     ASM_STORE16_REG_REG_OFFSET(emit->as, reg_value, reg_base, index_value);
                     #else
-                    #if N_XTENSA || N_XTENSAWIN
-                    if (index_value >= 0 && index_value < 256) {
-                        asm_xtensa_op_s16i(emit->as, reg_value, reg_base, index_value);
-                        break;
-                    }
-                    #endif
                     if (index_value != 0) {
                         // index is a non-zero immediate
                         ASM_MOV_REG_IMM(emit->as, reg_index, index_value << 1);