py/asmthumb: Extend load/store generators with ARMv7-M opcodes.

This commit lets the Thumb native code generator backend emit ARMv7-M
specific opcodes for indexed load/store operations if possible.

Now T3 opcode encodings are used if the generator backend is configured
to allow emitting ARMv7-M opcodes and if the (unsigned) scaled index
fits in 12 bits.  Or, in other words, LDR{B,H}.W and STR{B,H}.W opcodes
are now emitted if possible.

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

1 files changed, 16 insertions, 11 deletions

diff --git a/py/asmthumb.h b/py/asmthumb.h
index 570b44b96..6c83b583e 100644
--- a/py/asmthumb.h
+++ b/py/asmthumb.h
@@ -382,8 +382,10 @@ void asm_thumb_mov_reg_local(asm_thumb_t *as, uint rlo_dest, int local_num); //
 void asm_thumb_mov_reg_local_addr(asm_thumb_t *as, uint rlo_dest, int local_num); // convenience
 void asm_thumb_mov_reg_pcrel(asm_thumb_t *as, uint rlo_dest, uint label);
 
-void asm_thumb_ldr_reg_reg_i12_optimised(asm_thumb_t *as, uint reg_dest, uint reg_base, uint word_offset); // convenience
-void asm_thumb_ldrh_reg_reg_i12_optimised(asm_thumb_t *as, uint reg_dest, uint reg_base, uint uint16_offset); // convenience
+// Generate optimised load dest, [src, #offset] sequence
+void asm_thumb_load_reg_reg_offset(asm_thumb_t *as, uint reg_dest, uint reg_base, uint offset, uint shift);
+// Generate optimised store src, [dest, #offset] sequence
+void asm_thumb_store_reg_reg_offset(asm_thumb_t *as, uint reg_src, uint reg_base, uint offset, uint shift);
 
 void asm_thumb_b_label(asm_thumb_t *as, uint label); // convenience: picks narrow or wide branch
 void asm_thumb_bcc_label(asm_thumb_t *as, int cc, uint label); // convenience: picks narrow or wide branch
@@ -463,17 +465,20 @@ void asm_thumb_b_rel12(asm_thumb_t *as, int rel);
 #define ASM_MUL_REG_REG(as, reg_dest, reg_src) asm_thumb_format_4((as), ASM_THUMB_FORMAT_4_MUL, (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_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_LOAD16_REG_REG_OFFSET(as, reg_dest, reg_base, uint16_offset) asm_thumb_ldrh_reg_reg_i12_optimised((as), (reg_dest), (reg_base), (uint16_offset))
-#define ASM_LOAD32_REG_REG(as, reg_dest, reg_base) asm_thumb_ldr_rlo_rlo_i5((as), (reg_dest), (reg_base), 0)
-#define ASM_LOAD32_REG_REG_OFFSET(as, reg_dest, reg_base, word_offset) asm_thumb_ldr_reg_reg_i12_optimised((as), (reg_dest), (reg_base), (word_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_thumb_load_reg_reg_offset((as), (reg_dest), (reg_base), (byte_offset), 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_thumb_load_reg_reg_offset((as), (reg_dest), (reg_base), (halfword_offset), 1)
+#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_thumb_load_reg_reg_offset((as), (reg_dest), (reg_base), (word_offset), 2)
 
 #define ASM_STORE_REG_REG_OFFSET(as, reg_src, reg_base, word_offset) ASM_STORE32_REG_REG_OFFSET((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)
-#define ASM_STORE32_REG_REG(as, reg_src, reg_base) asm_thumb_str_rlo_rlo_i5((as), (reg_src), (reg_base), 0)
-#define ASM_STORE32_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_STORE8_REG_REG_OFFSET((as), (reg_src), (reg_base), 0)
+#define ASM_STORE8_REG_REG_OFFSET(as, reg_src, reg_base, byte_offset) asm_thumb_store_reg_reg_offset((as), (reg_src), (reg_base), (byte_offset), 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_thumb_store_reg_reg_offset((as), (reg_src), (reg_base), (halfword_offset), 1)
+#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_src, reg_base, word_offset) asm_thumb_store_reg_reg_offset((as), (reg_src), (reg_base), (word_offset), 2)
 
 #define ASM_LOAD8_REG_REG_REG(as, reg_dest, reg_base, reg_index) asm_thumb_ldrb_rlo_rlo_rlo((as), (reg_dest), (reg_base), (reg_index))
 #define ASM_LOAD16_REG_REG_REG(as, reg_dest, reg_base, reg_index) \