Teach top to merge or delete "addi sp, sp, X".

This reduces code size, because ncg emits too many "addi sp, sp, X"
instructions when unstacking things.  Now top lowers "addi sp, sp, X"
by lifting other instructions.  This sometimes creates chances to
merge or delete _addi_ instructions.  If no such chance is found, the
_addi_ remains uselessly lowered.

Edit ncg/table to remove something that top now does.

Edit ncg/mach.c to remove some spaces after commas.  This removes a
whitespace difference between *.s and *.so files, because top removes
the space.

mach/powerpc/ncg/table

@@ -334,9 +334,9 @@ INSTRUCTIONS
   lhax            GPR:wo, GPR:ro, GPR:ro cost(4, 3).
   lhz             GPR:wo, SET_RC_H:ro cost(4, 3).
   lhzx            GPR:wo, GPR:ro, GPR:ro cost(4, 3).
+  lwz             GPR+LOCAL:wo, SET_RC_W:ro cost(4, 3).
   lwzu            GPR:wo, IND_RC_W:rw cost(4, 3).
   lwzx            GPR:wo, GPR:ro, GPR:ro cost(4, 3).
-  lwz             GPR+LOCAL:wo, SET_RC_W:ro cost(4, 3).
   mfcr            GPR:wo cost(4,2).
   mfspr           GPR:wo, SPR:ro cost(4, 3).
   mtspr           SPR:wo, GPR:ro cost(4, 2).
@@ -361,7 +361,6 @@ INSTRUCTIONS
   rlwnm           GPR:wo, GPR:ro, GPR:ro, CONST:ro, CONST:ro.
     rotlw         GPR+LOCAL:wo, GPR:ro, GPR:ro.
   slw             GPR+LOCAL:wo, GPR:ro, GPR:ro.
-  subf            GPR:wo, GPR:ro, GPR:ro.
   sraw            GPR+LOCAL:wo, GPR:ro, GPR:ro cost(4, 2).
   srawi           GPR+LOCAL:wo, GPR:ro, CONST:ro cost(4, 2).
   srw             GPR+LOCAL:wo, GPR:ro, GPR:ro.
@@ -378,6 +377,7 @@ INSTRUCTIONS
   stw             GPR:ro, SET_RC_W:rw cost(4, 3).
   stwx            GPR:ro, GPR:ro, GPR:ro cost(4, 3).
   stwu            GPR:ro, IND_RC_W:rw cost(4, 3).
+  subf            GPR:wo, GPR:ro, GPR:ro.
   xor             GPR:wo, GPR:ro, GPR:ro.
   xori            GPR:wo, GPR:ro, CONST:ro.
   xoris           GPR:wo, GPR:ro, CONST:ro.
@@ -762,6 +762,10 @@ STACKINGRULES
 
 COERCIONS
 
+	/* The unstacking coercions emit many "addi sp, sp, X"
+	 * instructions; the target optimizer (top) will merge them.
+	 */
+
 	from STACK
 		uses REG
 		gen
@@ -2103,12 +2107,13 @@ PATTERNS
 			mr fp, r0
 			blr.
 
+	/* If "ret" coerces STACK to REG3, then top will delete the
+	 * extra "addi sp, sp, 4".
+	 */
+
 	pat ret $1==4                      /* Return from procedure, word */
 		with REG3
 			leaving ret 0
-		with STACK
-			gen lwz r3, {IND_RC_W, sp, 0}
-			leaving ret 0
 
 	pat ret $1==8                      /* Return from proc, double-word */
 		with REG3 INT_W
@@ -2117,11 +2122,6 @@ PATTERNS
 		with REG3 STACK
 			gen lwz r4, {IND_RC_W, sp, 0}
 			leaving ret 0
-		with STACK
-			gen
-				lwz r3, {IND_RC_W, sp, 0}
-				lwz r4, {IND_RC_W, sp, 4}
-			leaving ret 0
 
 	/*
 	 * These rules for blm/bls are wrong if length is zero.

mach/powerpc/top/table

@@ -1,11 +1,12 @@
 
 /* PowerPC table for ACK target optimizer */
 
-MAXOP 3;
+MAXOP 5;
 LABEL_STARTER '.';
 
 %%;
 
+L1, L2, L3, L4, L5  { not_using_sp(VAL) };
 RNZ                 { strcmp(VAL, "r0") };  /* not r0 */
 X, Y, Z             { TRUE };
 
@@ -16,6 +17,47 @@ X, Y, Z             { TRUE };
 addi  RNZ, RNZ, 0            -> ;
 addis RNZ, RNZ, 0            -> ;
 
+addi RNZ, RNZ, X : addi RNZ, RNZ, Y { plus(X, Y, Z) }
+                             -> addi RNZ, RNZ, Z ;
+
+/* Lower "addi sp, sp, X" by lifting other instructions, looking for
+ * chances to merge or delete _addi_ instructions, and assuming that
+ * the code generator uses "sp" not "r1".
+ */
+addi sp, sp, X : ANY L1                 { lift(ANY) }
+                             -> ANY L1                 : addi sp, sp, X ;
+addi sp, sp, X : ANY L1, L2             { lift(ANY) }
+                             -> ANY L1, L2             : addi sp, sp, X ;
+addi sp, sp, X : ANY L1, L2, L3         { lift(ANY) }
+                             -> ANY L1, L2, L3         : addi sp, sp, X ;
+addi sp, sp, X : ANY L1, L2, L3, L4     { lift(ANY) }
+ -> ANY L1, L2, L3, L4     : addi sp, sp, X ;
+addi sp, sp, X : ANY L1, L2, L3, L4, L5 { lift(ANY) }
+                             -> ANY L1, L2, L3, L4, L5 : addi sp, sp, X ;
+addi sp, sp, X : lmw Y, L1 { Y[0]=='r' && atoi(Y+1)>1 }
+                             -> lmw Y, L1 : addi sp, sp, X ;
+
+/* Merge _addi_ when popping from the stack. */
+addi sp, sp, X : lwz L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> lwz L1, Z(sp) : addi sp, sp, X ;
+addi sp, sp, X : lfs L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> lfs L1, Z(sp) : addi sp, sp, X ;
+addi sp, sp, X : lfd L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> lfd L1, Z(sp) : addi sp, sp, X ;
+
+/* Lower or delete _addi_ when pushing to the stack. */
+addi sp, sp, X : stwu  L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> stw  L1, Z(sp) : addi sp, sp, Z ;
+addi sp, sp, X : stfsu L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> stfs L1, Z(sp) : addi sp, sp, Z ;
+addi sp, sp, X : stfdu L1, Y(sp) { plus(X, Y, Z) && Z[0]!='-' }
+                             -> stfd L1, Z(sp) : addi sp, sp, Z ;
+addi sp, sp, 4 : stfdu L1, -8(sp) -> stfdu L1, -4(sp) ;
+
+/* Delete _addi_ when setting the stack pointer. */
+addi sp, sp, X : addi sp, L1, Y   -> addi sp, L1, Y ;
+addi sp, sp, X : lwz sp, L1       -> lwz sp, L1 ;
+
 or X, Y, Y                   -> mr X, Y ;
 or. X, Y, Y                  -> mr. X, Y ;
 
@@ -50,3 +92,89 @@ b X : labdef X               -> labdef X ;
 /* LT=0, GT=1, EQ=2, OV=3 */
 
 %%;
+
+/* Is it a word character? 0-9A-Za-z_ */
+static int isword(char c) {
+	return
+	    (c >= '0' && c <= '9') || (c >= 'A' && c <= 'Z') ||
+	    (c >= 'a' && c <= 'z') || (c == '_');
+}
+
+/* Does operand _s_ not use the stack pointer? */
+int not_using_sp(const char *s) {
+	int boundary;
+
+	boundary = 1;
+	while (*s) {
+		if (boundary &&
+		    ((s[0]=='s' && s[1]=='p') || (s[0]=='r' && s[1]=='1')) &&
+		    !isword(s[2]))
+			return 0;
+		boundary = !isword(*s);
+		s++;
+	}
+	return 1;
+}
+
+
+/* Instructions to lift(), sorted in strcmp() order.  These are from
+ * ../ncg/table, minus branch instructions.
+ */
+const char *liftables[] = {
+	"add", "add.", "addi",
+	"and", "andc", "andi.", "andis.",
+	"cmp", "cmpi", "cmpl", "cmpli",
+	"cmplw", "cmplwi", "cmpw", "cmpwi",
+	"divw", "divwu", "eqv", "extlwi", "extrwi", "extsb", "extsh",
+	"fadd", "fadds", "fcmpo", "fctiwz", "fdiv", "fdivs",
+	"fmr", "fmul", "fmuls", "fneg", "frsp", "fsub", "fsubs",
+	"lbz", "lbzx",
+	"lfd", "lfdu", "lfdx", "lfs", "lfsu", "lfsx",
+	"lha", "lhax", "lhz", "lhzx",
+	"li", "lis", "lwz", "lwzu", "lwzx",
+	"mfcr", "mfspr", "mr", "mr.", "mtspr", "mullw",
+	"nand", "neg", "nor", "or", "or.", "ori", "oris",
+	"rlwinm", "rlwnm", "rotlwi", "rotrwi",
+	"slw", "slwi", "sraw", "srawi", "srw", "srwi",
+	"stb", "stbx",
+	"stfd", "stfdu", "stfdx", "stfs", "stfsu", "stfsx",
+	"sth", "sthx", "stw", "stwx", "stwu",
+	"subf", "xor", "xori", "xoris",
+};
+
+static int liftcmp(const void *a, const void *b) {
+	return strcmp(*(const char **)a, *(const char **)b);
+}
+
+/* May we lift instruction _s_ above "addi SP, SP, X"? */
+int lift(const char *s) {
+	return bsearch(&s, liftables,
+	    sizeof(liftables) / sizeof(liftables[0]),
+	    sizeof(liftables[0]), liftcmp);
+}
+
+
+/* Does it fit a signed 16-bit integer? */
+static int fits16(long l) {
+	return l >= -32768 && l <= 32767;
+}
+
+/* Tries sum = a + b with signed 16-bit integers. */
+int plus(const char *a, const char *b, const char *sum)
+{
+	long la, lb, lsum;
+	char *end;
+
+	la = strtol(a, &end, 10);
+	if (*a == '\0' || *end != '\0' || !fits16(la))
+		return 0;
+	lb = strtol(b, &end, 10);
+	if (*b == '\0' || *end != '\0' || !fits16(lb))
+		return 0;
+
+	lsum = la + lb;
+	if (!fits16(lsum))
+		return 0;
+	snprintf(sum, 7, "%ld", lsum);
+	return 1;
+}