Merge pull request #15 from catseye/16-bit-compare

16-bit compare instruction

HISTORY.md

@@ -4,6 +4,8 @@ History of SixtyPical
 0.18
 ----
 
+*   `cmp` instruction can now perform a 16-bit unsigned comparison
+    of `word` memory locations (at the cost of trashing `a`.)
 *   Fixed pathological memory use in the lexical scanner - should
     be much less inefficient now when parsing large source files.
 

TODO.md

@@ -1,17 +1,6 @@
 TODO for SixtyPical
 ===================
 
-### 16-bit `cmp`
-
-This is because we don't actually want `low` and `high` address operators
-that turn `word` type into `byte`.
-
-This is because this immediately makes things harder (that is, effectively
-impossible) to analyze.
-
-16-bit `cmp` also benefits from some special differences between `cmp`
-and `sub` on 6502, so it would be nice to capture them.
-
 ### Save values to other-than-the-stack
 
 Allow

doc/SixtyPical.md

@@ -288,9 +288,9 @@ this mode is used.
 
     copy <src-memory-location>, <dest-memory-location>
 
-Reads from src and writes to dest.  Differs from `st` in that is able to
-copy more general types of data (for example, vectors,) and it trashes the
-`z` and `n` flags and the `a` register.
+Reads from src and writes to dest.  Differs from `ld` and `st` in that
+it is able to copy more general types of data (for example, vectors,)
+and it trashes the `z` and `n` flags and the `a` register.
 
 *   It is illegal if dest is read-only.
 *   It is illegal if dest does not occur in the WRITES of the current routine.
@@ -376,6 +376,9 @@ and initializing them afterwards.
 
 dest and src continue to be initialized afterwards.
 
+In addition, if dest is of `word` type, then src must also be of `word`
+type, and in this case this instruction trashes the `a` register.
+
 ### dec ###
 
     dec <dest-memory-location>
@@ -395,12 +398,24 @@ and initializing them afterwards.
 
 Subtracts the contents of src from dest (without considering carry) but
 does not store the result anywhere, only sets the resulting flags.
+This means that `z` is set if src and dest are equal,
+and `c` is set if dest is greater than or equal to src
+(`c` is unset if dest is less than src.)
 
 *   It is illegal if src OR dest is uninitialized.
 
 Affects n, z, and c flags, requiring that they be in the WRITES,
 and initializing them afterwards.
 
+In addition, if dest is of `word` type, then src must also be of `word`
+type, and in this case this instruction trashes the `a` register.
+
+Note that `cmp` is not suitable for making a
+signed comparison; this article, which mentions
+techniques that a SixtyPical compiler could use to
+implement `cmp`, also explains why that is:
+[Beyond 8-bit Unsigned Comparisons, by Bruce Clark](http://www.6502.org/tutorials/compare_beyond.html).
+
 ### and, or, xor ###
 
     and <dest-memory-location>, <src-memory-location>

eg/rudiments/cmp-byte.60p

@@ -0,0 +1,74 @@
+// Should print ENGGL
+
+byte b
+
+define chrout routine
+  inputs a
+  trashes a
+  @ 65490
+
+define main routine
+  outputs b
+  trashes a, x, y, z, n, c, v
+{
+    ld a, 40
+    st a, b
+
+    cmp a, b
+    if z {
+        ld a, 69  // E
+        call chrout
+    } else {
+        ld a, 78  // N
+        call chrout
+    }
+
+    ld a, 41
+    st a, b
+    ld a, 40
+
+    cmp a, b
+    if z {
+        ld a, 69  // E
+        call chrout
+    } else {
+        ld a, 78  // N
+        call chrout
+    }
+
+    ld a, 20
+    st a, b
+
+    ld a, 21
+
+    cmp a, b      // 21 >= 20
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+
+    ld a, 20
+
+    cmp a, b      // 20 >= 20
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+
+    ld a, 19
+
+    cmp a, b      // 19 < 20
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+}

eg/rudiments/cmp-word.60p

@@ -0,0 +1,72 @@
+// Should print ENGGL
+
+word w1
+word w2
+
+define chrout routine
+  inputs a
+  trashes a
+  @ 65490
+
+define main routine
+  outputs w1, w2
+  trashes a, x, y, z, n, c, v
+{
+    copy 4000, w1
+    copy 4000, w2
+
+    cmp w1, w2
+    if z {
+        ld a, 69  // E
+        call chrout
+    } else {
+        ld a, 78  // N
+        call chrout
+    }
+
+    copy 4000, w1
+    copy 4001, w2
+
+    cmp w1, w2
+    if z {
+        ld a, 69  // E
+        call chrout
+    } else {
+        ld a, 78  // N
+        call chrout
+    }
+
+    copy 20002, w1
+    copy 20001, w2
+
+    cmp w1, w2    // 20002 >= 20001
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+
+    copy 20001, w1
+
+    cmp w1, w2    // 20001 >= 20001
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+
+    copy 20000, w1
+
+    cmp w1, w2    // 20000 < 20001
+    if c {
+        ld a, 71  // G
+        call chrout
+    } else {
+        ld a, 76  // L
+        call chrout
+    }
+}

src/sixtypical/analyzer.py

@@ -505,8 +505,12 @@ class Analyzer(object):
             context.assert_meaningful(src, dest)
             if isinstance(src, IndexedRef):
                 context.assert_types_for_read_table(instr, src, dest, TYPE_BYTE)
-            else:
+            elif src.type == TYPE_BYTE:
                 self.assert_type(TYPE_BYTE, src, dest)
+            else:
+                self.assert_type(TYPE_WORD, src, dest)
+                context.set_touched(REG_A)
+                context.set_unmeaningful(REG_A)
             context.set_written(FLAG_Z, FLAG_N, FLAG_C)
         elif opcode == 'and':
             if isinstance(src, IndexedRef):

src/sixtypical/compiler.py

@@ -391,6 +391,17 @@ class Compiler(object):
 
     def compile_cmp(self, instr, src, dest):
         """`instr` is only for reporting purposes"""
+        if isinstance(src, LocationRef) and src.type == TYPE_WORD:
+            src_label = self.get_label(src.name)
+            dest_label = self.get_label(dest.name)
+            self.emitter.emit(LDA(Absolute(dest_label)))
+            self.emitter.emit(CMP(Absolute(src_label)))
+            end_label = Label('end_label')
+            self.emitter.emit(BNE(Relative(end_label)))
+            self.emitter.emit(LDA(Absolute(Offset(dest_label, 1))))
+            self.emitter.emit(CMP(Absolute(Offset(src_label, 1))))
+            self.emitter.resolve_label(end_label)
+            return
         cls = {
             'a': CMP,
             'x': CPX,

tests/SixtyPical Analysis.md

@@ -1118,6 +1118,52 @@ Some rudimentary tests for `cmp`.
     | }
     ? UnmeaningfulReadError: a
 
+`cmp` can work on words. In this case, it trashes `a`.
+
+    | word za
+    | word zb
+    | 
+    | define main routine
+    |   inputs za, zb
+    |   trashes a, z, c, n
+    | {
+    |     cmp za, zb
+    | }
+    = ok
+
+    | word za
+    | word zb
+    | 
+    | define main routine
+    |   inputs za, zb
+    |   trashes a, z, n
+    | {
+    |     cmp za, zb
+    | }
+    ? ForbiddenWriteError: c
+
+    | word za
+    | word zb
+    | 
+    | define main routine
+    |   inputs za, zb
+    |   trashes z, c, n
+    | {
+    |     cmp za, zb
+    | }
+    ? ForbiddenWriteError: a
+
+    | word za
+    | word zb
+    | 
+    | define main routine
+    |   inputs za
+    |   trashes z, c, n
+    | {
+    |     cmp za, zb
+    | }
+    ? UnmeaningfulReadError: zb
+
 ### and ###
 
 Some rudimentary tests for `and`.

tests/SixtyPical Compilation.md

@@ -385,6 +385,26 @@ Some instructions on tables. (3/3)
     = $081B   DEC $081F,X
     = $081E   RTS
 
+Compiling 16-bit `cmp`.
+
+    | word za @ 60001
+    | word zb : 3003
+    | 
+    | define main routine
+    |   inputs za, zb
+    |   trashes a, z, c, n
+    | {
+    |     cmp za, zb
+    | }
+    = $080D   LDA $EA61
+    = $0810   CMP $081C
+    = $0813   BNE $081B
+    = $0815   LDA $EA62
+    = $0818   CMP $081D
+    = $081B   RTS
+    = $081C   .byte $BB
+    = $081D   .byte $0B
+
 Compiling `if`.
 
     | define main routine