py: Big improvements to inline assembler.

Improved the Thumb assembler back end.  Added many more Thumb
instructions to the inline assembler.  Improved parsing of assembler
instructions and arguments.  Assembler functions can now be passed the
address of any object that supports the buffer protocol (to get the
address of the buffer).  Added an example of how to sum numbers from
an array in assembler.

2 files changed, 76 insertions, 15 deletions

diff --git a/examples/asmled.py b/examples/asmled.py
index e0d6c73ce..917d9ba03 100644
--- a/examples/asmled.py
+++ b/examples/asmled.py
@@ -1,4 +1,5 @@
 # flash LED #1 using inline assembler
+# this version is overly verbose and uses word stores
 @micropython.asm_thumb
 def flash_led(r0):
     movw(r1, (stm.GPIOA + stm.GPIO_BSRRL) & 0xffff)
@@ -13,69 +14,72 @@ def flash_led(r0):
     label(loop1)
 
     # turn LED on
-    str(r2, r1, 0)
+    str(r2, [r1, 0])
 
     # delay for a bit
     movw(r4, 5599900 & 0xffff)
     movt(r4, (5599900 >> 16) & 0xffff)
     label(delay_on)
-    subs(r4, r4, 1)
+    sub(r4, r4, 1)
     cmp(r4, 0)
     bgt(delay_on)
 
     # turn LED off
-    str(r3, r1, 0)
+    str(r3, [r1, 0])
 
     # delay for a bit
     movw(r4, 5599900 & 0xffff)
     movt(r4, (5599900 >> 16) & 0xffff)
     label(delay_off)
-    subs(r4, r4, 1)
+    sub(r4, r4, 1)
     cmp(r4, 0)
     bgt(delay_off)
 
     # loop r0 times
-    subs(r0, r0, 1)
+    sub(r0, r0, 1)
     label(loop_entry)
     cmp(r0, 0)
     bgt(loop1)
 
-# flash LED #1 using inline assembler
-# this version uses the convenience assembler operation 'movwt'
+# flash LED #2 using inline assembler
+# this version uses half-word sortes, and the convenience assembler operation 'movwt'
 @micropython.asm_thumb
 def flash_led_v2(r0):
-    movwt(r1, stm.GPIOA + stm.GPIO_BSRRL)
-    movwt(r2, 1 << 13)
-    movwt(r3, 1 << (16 + 13))
+    # get the GPIOA address in r1
+    movwt(r1, stm.GPIOA)
+
+    # get the bit mask for PA14 (the pin LED #2 is on)
+    movw(r2, 1 << 14)
 
     b(loop_entry)
 
     label(loop1)
 
     # turn LED on
-    str(r2, r1, 0)
+    strh(r2, [r1, stm.GPIO_BSRRL])
 
     # delay for a bit
     movwt(r4, 5599900)
     label(delay_on)
-    subs(r4, r4, 1)
+    sub(r4, r4, 1)
     cmp(r4, 0)
     bgt(delay_on)
 
     # turn LED off
-    str(r3, r1, 0)
+    strh(r2, [r1, stm.GPIO_BSRRH])
 
     # delay for a bit
     movwt(r4, 5599900)
     label(delay_off)
-    subs(r4, r4, 1)
+    sub(r4, r4, 1)
     cmp(r4, 0)
     bgt(delay_off)
 
     # loop r0 times
-    subs(r0, r0, 1)
+    sub(r0, r0, 1)
     label(loop_entry)
     cmp(r0, 0)
     bgt(loop1)
 
+flash_led(5)
 flash_led_v2(5)
diff --git a/examples/asmsum.py b/examples/asmsum.py
new file mode 100644
index 000000000..07e71c738
--- /dev/null
+++ b/examples/asmsum.py
@@ -0,0 +1,57 @@
+@micropython.asm_thumb
+def asm_sum_words(r0, r1):
+
+    # r0 = len
+    # r1 = ptr
+    # r2 = sum
+    # r3 = dummy
+    mov(r2, 0)
+
+    b(loop_entry)
+
+    label(loop1)
+    ldr(r3, [r1, 0])
+    add(r2, r2, r3)
+
+    add(r1, r1, 4)
+    sub(r0, r0, 1)
+
+    label(loop_entry)
+    cmp(r0, 0)
+    bgt(loop1)
+
+    mov(r0, r2)
+
+@micropython.asm_thumb
+def asm_sum_bytes(r0, r1):
+
+    # r0 = len
+    # r1 = ptr
+    # r2 = sum
+    # r3 = dummy
+    mov(r2, 0)
+
+    b(loop_entry)
+
+    label(loop1)
+    ldrb(r3, [r1, 0])
+    add(r2, r2, r3)
+
+    add(r1, r1, 1)
+    sub(r0, r0, 1)
+
+    label(loop_entry)
+    cmp(r0, 0)
+    bgt(loop1)
+
+    mov(r0, r2)
+
+import array
+
+b = array.array('l', (100, 200, 300, 400))
+n = asm_sum_words(len(b), b)
+print(b, n)
+
+b = array.array('b', (10, 20, 30, 40, 50, 60, 70, 80))
+n = asm_sum_bytes(len(b), b)
+print(b, n)