Add a 'kill' pseudoinstruction to the 8080 assembler, which marks when the code

generator is finished with a register; use this to create some more effective
peephole optimiser rules.

mach/i80/as/mach2.c

@@ -21,3 +21,5 @@
 %token <y_word> STLDAX
 
 %type <y_word> r16 r8
+
+%token <y_word> KILL

mach/i80/as/mach3.c

@@ -4,6 +4,12 @@
  */
 #define RCSID3 "$Id$"
 
+/*
+ * Specials
+ */
+
+0,  KILL,       0,      "kill",
+
 /*
  * Intel 8080 keywords
  */

mach/i80/as/mach4.c

@@ -10,7 +10,8 @@
  * Intel 8080 parsing rules
  */
 operation
-	:	NOOPOP
+	:   KILL REG {}
+	|	NOOPOP
 			{	emit1($1);}
 	|
 		D8OP expr

mach/i80/ncg/table

@@ -48,6 +48,7 @@ TOKENS
 
 const1			= { INT num; } 1 num.
 const2			= { INT num; } 2 num.
+largeconst2    = { INT num; } 2 num. /* Word-sized constant with low byte zero */
 smallpconst2   = { INT num; } 2 num. /* Byte-sized positive constant */
 smallnconst2   = { INT num; } 2 num. /* Byte-sized negative constant */
 label			= { ADDR off; } 2 off.
@@ -60,17 +61,18 @@ SETS
 reg1			= reg + lbreg + m.
 b_d_h_sp		= regpair + stackpointer.
 b_d_h_psw		= regpair + psword.
-immediate		= smallpconst2 + smallnconst2 + const2 + label.
+immediate		= largeconst2 + smallpconst2 + smallnconst2 + const2 + label.
 src1			= reg.
 src2			= hl_or_de + const2 + label.
 src1or2			= src1 + src2.
+anyreg      = reg + regpair.
 
 INSTRUCTIONS
 
 /* aci	const1:ro	kills a:cc	cost(2, 7).	*/
    adc	reg1:ro		kills a:cc	cost(1, 4).
    add	reg1:ro 	kills a:cc	cost(1, 4).
-/* adi	const1:ro 	kills a:cc	cost(2, 7).	*/
+   adi	const1:ro 	kills a:cc	cost(2, 7).
    ana	reg1:ro 	kills a:cc	cost(1, 4).
    ani	const1:ro 	kills a:cc	cost(2, 7).
    Call	"call" label:ro			cost(3,17).
@@ -154,6 +156,7 @@ INSTRUCTIONS
    xri	const1:ro 	kills a:cc	cost(2, 7).
    xthl			kills hl	cost(1,18).
    
+   kill anyreg:rw kills :cc cost(0, 0).
 
 MOVES
 
@@ -261,6 +264,9 @@ COERCIONS
 
    from hl_or_de				yields %1.2
 
+   from largeconst2
+      yields {const2, %1.num}
+
    from smallpconst2
       yields {const2, %1.num}
 
@@ -271,6 +277,10 @@ COERCIONS
       uses hl_or_de=%1
       yields %a
 
+   from largeconst2
+      uses hl_or_de=%1
+      yields %a
+      
    from smallpconst2
       uses reg={const1, %1.num & 0xff}
       yields %a
@@ -305,6 +315,9 @@ PATTERNS
 /* Group 1: Load instructions                */
 /*********************************************/
 
+pat loc ($1 != 0) && (($1 & 0xff) == 0)
+   yields {largeconst2, $1}
+
 pat loc ufit($1, 8)
    yields {smallpconst2, $1}
 
@@ -727,6 +740,18 @@ pat sde
 /****************************************/
 
 pat adi $1==2
+   with hl_or_de largeconst2
+      uses areg
+      gen
+         mov a, %1.1
+         adi {const1, %2.num >> 8}
+         mov %1.1, a
+         kill a
+      yields %1
+   with largeconst2 hl_or_de
+      yields %1 %2
+      leaving
+         adi 2
    with hlreg dereg
       gen
          dad de
@@ -755,6 +780,14 @@ pat sbi $1==2
       yields %2 {const2, 0-%1.num}
       leaving
          adi 2
+   with largeconst2 hl_or_de
+      yields %2 {largeconst2, 0-%1.num}
+      leaving
+         adi 2
+   with smallpconst2 hl_or_de
+      yields %2 {smallnconst2, 0-%1.num}
+      leaving
+         adi 2
    with smallnconst2 hl_or_de
       yields %2 {smallpconst2, 0-%1.num}
       leaving
@@ -768,6 +801,7 @@ pat sbi $1==2
          mvi a, {const1, %2.num >> 8}
          sbb %1.1
          mov %1.1, a
+         kill a
       yields %1
    with hl_or_de hl_or_de
       uses areg
@@ -778,6 +812,7 @@ pat sbi $1==2
          mov a,%2.1
          sbb %1.1
          mov %1.1,a
+         kill a
       yields %1
    with hl_or_de hl_or_de
       uses areg
@@ -788,6 +823,7 @@ pat sbi $1==2
          mov a,%2.1
          sbb %1.1
          mov %2.1,a
+         kill a
       yields %2
 
 pat sbi $1==4
@@ -1282,6 +1318,7 @@ pat and $1==2
          mov a, %1.2
          ani {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
       yields %1
    with smallnconst2 hl_or_de
       yields %1 %2
@@ -1293,9 +1330,11 @@ pat and $1==2
          mov a, %1.2
          ani {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
          mov a, %1.1
          ani {const1, %2.num >> 8}
          mov %1.1, a
+         kill a
       yields %1
    with const2 hl_or_de
       yields %1 %2
@@ -1307,9 +1346,11 @@ pat and $1==2
          mov a, %1.2
          ana %2.2
          mov %2.2, a
+         kill a
          mov a, %1.1
          ana %2.1
          mov %2.1, a
+         kill a
       yields %2
    with hl_or_de hl_or_de
       uses areg
@@ -1317,9 +1358,11 @@ pat and $1==2
          mov a,%1.2
          ana %2.2
          mov %1.2,a
+         kill a
          mov a,%1.1
          ana %2.1
          mov %1.1,a
+         kill a
       yields %1
 
 pat and defined($1)
@@ -1343,6 +1386,7 @@ pat ior $1==2
          mov a, %1.2
          ori {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
       yields %1
    with smallpconst2 hl_or_de
       yields %1 %2
@@ -1354,6 +1398,7 @@ pat ior $1==2
          mov a, %1.2
          ori {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
          move {const1, 0xff}, %1.1
       yields %1
    with smallnconst2 hl_or_de
@@ -1366,9 +1411,11 @@ pat ior $1==2
          mov a, %1.2
          ori {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
          mov a, %1.1
          ori {const1, %2.num >> 8}
          mov %1.1, a
+         kill a
       yields %1
    with const2 hl_or_de
       yields %1 %2
@@ -1380,9 +1427,11 @@ pat ior $1==2
          mov a, %1.2
          ora %2.2
          mov %2.2, a
+         kill a
          mov a, %1.1
          ora %2.1
          mov %2.1, a
+         kill a
       yields %2
    with hl_or_de hl_or_de
       uses areg
@@ -1390,9 +1439,11 @@ pat ior $1==2
          mov a,%1.2
          ora %2.2
          mov %1.2,a
+         kill a
          mov a,%1.1
          ora %2.1
          mov %1.1,a
+         kill a
       yields %1
 
 pat ior defined($1)
@@ -1414,6 +1465,7 @@ pat xor $1==2
          mov a, %1.2
          xri {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
       yields %1
    with smallpconst2 hl_or_de
       yields %1 %2
@@ -1425,9 +1477,11 @@ pat xor $1==2
          mov a, %1.2
          xri {const1, %2.num & 0xff}
          mov %1.2, a
+         kill a
          mov a, %1.1
          xri {const1, %2.num >> 8}
          mov %1.1, a
+         kill a
       yields %1
    with const2 hl_or_de
       yields %1 %2
@@ -1439,9 +1493,11 @@ pat xor $1==2
          mov a, %1.2
          xra %2.2
          mov %2.2, a
+         kill a
          mov a, %1.1
          xra %2.1
          mov %2.1, a
+         kill a
       yields %2
    with hl_or_de hl_or_de
       uses areg
@@ -1449,9 +1505,11 @@ pat xor $1==2
          mov a,%1.2
          xra %2.2
          mov %1.2,a
+         kill a
          mov a,%1.1
          xra %2.1
          mov %1.1,a
+         kill a
       yields %1
 
 pat xor defined($1)
@@ -1470,6 +1528,7 @@ uses areg
 gen mov a,%1.2
     cma.
     mov %1.2,a
+    kill a
     mov a,%1.1
     cma.
     mov %1.1,a				yields %1

mach/i80/top/table

@@ -26,4 +26,37 @@ push d : lxi d, X : pop h    -> lxi h, X : xchg ;
 
 push h : lhld X : pop d      -> xchg : lhld X ;
 
+X -1                         -> X 255 ;
+X -2                         -> X 254 ;
+X -3                         -> X 253 ;
+X -4                         -> X 252 ;
+
+mov a, X : adi -4 : mov X, a : kill a -> dcr X : dcr X : dcr X : dcr X;
+mov a, X : adi -3 : mov X, a : kill a -> dcr X : dcr X : dcr X ;
+mov a, X : adi -2 : mov X, a : kill a -> dcr X : dcr X ;
+mov a, X : adi -1 : mov X, a : kill a -> dcr X ;
+mov a, X : adi 0 : mov X, a : kill a -> ;
+mov a, X : adi 1 : mov X, a : kill a -> inr X ;
+mov a, X : adi 2 : mov X, a : kill a -> inr X : inr X ;
+mov a, X : adi 3 : mov X, a : kill a -> inr X : inr X : inr X ;
+mov a, X : adi 4 : mov X, a : kill a -> inr X : inr X : inr X : inr X;
+
+mov a, X : sbi -4 : mov X, a : kill a -> inr X : inr X : inr X : inr X;
+mov a, X : sbi -3 : mov X, a : kill a -> inr X : inr X : inr X ;
+mov a, X : sbi -2 : mov X, a : kill a -> inr X : inr X ;
+mov a, X : sbi -1 : mov X, a : kill a -> inr X ;
+mov a, X : sbi 0 : mov X, a : kill a -> ;
+mov a, X : sbi 1 : mov X, a : kill a -> dcr X ;
+mov a, X : sbi 2 : mov X, a : kill a -> dcr X : dcr X ;
+mov a, X : sbi 3 : mov X, a : kill a -> dcr X : dcr X : dcr X ;
+mov a, X : sbi 4 : mov X, a : kill a -> dcr X : dcr X : dcr X : dcr X ;
+
+mov a, X : ani 0 : mov X, a : kill a -> mvi X, 0 ;
+mov a, X : ani 255 : mov X, a : kill a -> ;
+
+mov a, X : ori 0 : mov X, a : kill a -> ;
+mov a, X : ori 255 : mov X, a : kill a -> mvi X, 255 ;
+
+mov a, X : xri 0 : mov X, a : kill a -> ;
+
 %%;

tests/plat/core/and_e.e

@@ -111,5 +111,19 @@ big
     asp 4
 7
 
+	/* And var with big low-byte-zero const */
+
+	loe big
+	loc 256
+	and EM_WSIZE
+	loc 256
+	cmu EM_WSIZE
+	zeq *8
+
+    loc __LINE__
+    cal $fail
+    asp 4
+8
+
     cal $finished
     end

tests/plat/core/intadd_e.c

@@ -27,5 +27,7 @@ void _m_a_i_n(void)
     ASSERT(((unsigned int)1   + (unsigned int)two) == 3);
     ASSERT(((unsigned int)-1  + (unsigned int)two) == 1);
 
+    ASSERT(((unsigned int)two + (unsigned int)256) == 258);
+
     finished();
 }

tests/plat/core/intsub_e.c

@@ -6,6 +6,7 @@ int two = 2;
 int one = 1;
 int zero = 0;
 int minusone = -1;
+int biggish = 258;
 
 /* Bypasses the CRT, so there's no stdio. */
 void _m_a_i_n(void)
@@ -28,5 +29,7 @@ void _m_a_i_n(void)
     ASSERT(((unsigned int)2   - (unsigned int)one) == 1);
     ASSERT(((unsigned int)1   - (unsigned int)two) == UINT_MAX);
 
+    ASSERT(((unsigned int)biggish - (unsigned int)256) == 2);
+
     finished();
 }

tests/plat/core/ior_e.e

@@ -111,5 +111,19 @@ big
     asp 4
 7
 
+	/* Or var with big low-byte-zero const */
+
+	loe big
+	loc 256
+	ior EM_WSIZE
+	loc 258
+	cmu EM_WSIZE
+	zeq *8
+
+    loc __LINE__
+    cal $fail
+    asp 4
+8
+
     cal $finished
     end

tests/plat/core/xor_e.e

@@ -111,6 +111,20 @@ big
     asp 4
 7
 
+	/* Xor var with big low-byte-zero const */
+
+	loe big
+	loc 256
+	xor EM_WSIZE
+	loc 745
+	cmu EM_WSIZE
+	zeq *8
+
+    loc __LINE__
+    cal $fail
+    asp 4
+8
+
 
     cal $finished
     end