Merge pull request #28 from kernigh/pr-set-ppc

add Modula-2 set operations to PowerPC ncg

mach/powerpc/libem/and.s

@@ -0,0 +1,24 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Set intersection.
+!  Stack: ( b a -- a*b )
+!  With r3 = size of set
+
+.define .and
+.and:
+	mr	r4, sp			! r4 = ptr to set a
+	add	r5, sp, r3		! r5 = ptr to set b
+	rlwinm	r6, r3, 30, 2, 31
+	mtspr	ctr, r6			! ctr = r3 / 4
+1:
+	lwz	r7, 0(r4)
+	lwz	r8, 0(r5)
+	and	r8, r7, r8		! intersection of words
+	stw	r8, 0(r5)
+	addi	r4, r4, 4
+	addi	r5, r5, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	add	sp, sp, r3
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/build.lua

@@ -7,7 +7,7 @@ for _, plat in ipairs(vars.plats) do
 	acklibrary {
 		name = "lib_"..plat,
 		srcs = {
-			"./*.s",
+			"./*.s", -- zer.s
 			"./*.e",
 		},
 		vars = { plat = plat },

mach/powerpc/libem/cms.s

@@ -0,0 +1,32 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Compare sets a, b.
+!  Stack: ( b a -- )
+!  With r3 = size of each set
+!  Yields r3 = 0 if equal, nonzero if not equal
+
+.define .cms
+.cms:
+	mr	r4, sp			! r4 = ptr to set a
+	add	r5, sp, r3		! r5 = ptr to set b
+	mr	r6, r3			! r6 = size
+	rlwinm	r3, r3, 30, 2, 31
+	mtspr	ctr, r3			! ctr = size / 4
+1:
+	lwz	r7, 0(r4)
+	lwz	r8, 0(r5)
+	cmp	cr0, 0, r7, r8		! compare words in sets
+	addi	r4, r4, 4
+	addi	r5, r5, 4
+	bc	IFFALSE, EQ, 2f		! branch if not equal
+	bc	DNZ, 0, 1b		! loop ctr times
+	addi	r3, r0, 0		! equal: return 0
+	b	3f
+2:
+	addi	r3, r0, 1		! not equal: return 1
+3:
+	rlwinm	r6, r6, 1, 0, 30	! r6 = size * 2
+	add	sp, sp, r6		! remove sets from stack
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/com.s

@@ -0,0 +1,20 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Set complement.
+!  Stack: ( a -- ~a )
+!  With r3 = size of set
+
+.define .com
+.com:
+	mr	r4, sp			! r4 = pointer to set a
+	rlwinm	r5, r3, 30, 2, 31
+	mtspr	ctr, r5			! ctr = r3 / 4
+1:
+	lwz	r6, 0(r4)
+	nor	r6, r6, r6		! complement of word
+	stw	r6, 0(r4)
+	addi	r4, r4, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/ior.s

@@ -0,0 +1,24 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Set union.
+!  Stack: ( b a -- a+b )
+!  With r3 = size of set
+
+.define .ior
+.ior:
+	mr	r4, sp			! r4 = ptr to set a
+	add	r5, sp, r3		! r5 = ptr to set b
+	rlwinm	r6, r3, 30, 2, 31
+	mtspr	ctr, r6			! ctr = r3 / 4
+1:
+	lwz	r7, 0(r4)
+	lwz	r8, 0(r5)
+	or	r8, r7, r8		! union of words
+	stw	r8, 0(r5)
+	addi	r4, r4, 4
+	addi	r5, r5, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	add	sp, sp, r3
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/set.s

@@ -0,0 +1,29 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Create singleton set.
+!  Stack: ( -- set )
+!  With r3 = size of set, r4 = bit number
+
+.define .set
+.set:
+	rlwinm	r7, r3, 30, 2, 31
+	neg	r5, r3
+	add	sp, sp, r5		! allocate set
+	mr	r6, sp			! r6 = ptr to set
+	mtspr	ctr, r7			! ctr = r3 / 4
+1:
+	rlwinm.	r7, r4, 0, 0, 26	! r7 = r4 & ~31
+	bc	IFTRUE, EQ, 2f		! branch if r4 in 0..31
+	addi	r5, r0, 0		! no bit, word is zero
+	b	3f
+2:
+	addi	r5, r0, 1
+	slw	r5, r5, r4		! yes bit, set bit in word
+3:
+	stw	r5, 0(r6)		! store word in set
+	addi	r4, r4, -32
+	addi	r6, r6, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/xor.s

@@ -0,0 +1,24 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Set symmetric difference.
+!  Stack: ( b a -- a/b )
+!  With r3 = size of set
+
+.define .xor
+.xor:
+	mr	r4, sp			! r4 = ptr to set a
+	add	r5, sp, r3		! r5 = ptr to set b
+	rlwinm	r6, r3, 30, 2, 31
+	mtspr	ctr, r6			! ctr = r3 / 4
+1:
+	lwz	r7, 0(r4)
+	lwz	r8, 0(r5)
+	xor	r8, r7, r8		! symmetric difference of words
+	stw	r8, 0(r5)
+	addi	r4, r4, 4
+	addi	r5, r5, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	add	sp, sp, r3
+	bclr	ALWAYS, 0, 0

mach/powerpc/libem/zer.s

@@ -0,0 +1,21 @@
+#include "powerpc.h"
+
+.sect .text
+
+! Create empty set.
+!  Stack: ( -- set )
+!  With r3 = size of set
+
+.define .zer
+.zer:
+	rlwinm	r7, r3, 30, 2, 31
+	addi	r4, r0, 0		! r4 = zero
+	neg	r5, r3
+	add	sp, sp, r5		! allocate set
+	mr	r6, sp			! r6 = ptr to set
+	mtspr	ctr, r7			! ctr = r3 / 4
+1:
+	stw	r4, 0(r6)		! store zero in set
+	addi	r6, r6, 4
+	bc	DNZ, 0, 1b		! loop ctr times
+	bclr	ALWAYS, 0, 0

mach/powerpc/ncg/table

@@ -10,7 +10,7 @@ INT64 = 8
 FP_OFFSET = 0   /* Offset of saved FP relative to our FP */
 PC_OFFSET = 4   /* Offset of saved PC relative to our FP */
 
-#define COMMENT(n) /* noop */
+#define COMMENT(n) /* comment {LABEL, n} */
 
 
 #define nicesize(x) ((x)==INT8 || (x)==INT16 || (x)==INT32 || (x)==INT64)
@@ -389,7 +389,7 @@ MOVES
 
 /* Constants */
 
-	from CONST_ALL smalls(%val) to GPR
+	from CONST_ALL + CONST smalls(%val) to GPR
 		gen
 			COMMENT("move CONST_ALL->GPR smalls")
 			addi %2, R0, {CONST, %1.val}
@@ -1160,7 +1160,7 @@ PATTERNS
 			loc $1
 			los INT32
 
-	pat los                            /* Load arbitrary size */
+	pat los $1==INT32                  /* Load arbitrary size */
 		with GPR3 GPR4 STACK
 			kills ALL
 			gen
@@ -1283,7 +1283,7 @@ PATTERNS
 			loc $1
 			sts INT32
 
-	pat sts                            /* Store arbitrary size */
+	pat sts $1==INT32                  /* Store arbitrary size */
 		with GPR3 GPR4 STACK
 			kills ALL
 			gen
@@ -1463,9 +1463,11 @@ PATTERNS
 				andisX %a, %2, {CONST, hi(%1.val)}
 			yields %a
 
-	pat and !defined($1)               /* AND set */
+	pat and defined($1)                /* AND set */
 		with STACK
+			kills ALL
 			gen
+				move {CONST, $1}, R3
 				bl {LABEL, ".and"}
 
 	pat ior $1==4                      /* OR word */
@@ -1498,8 +1500,17 @@ PATTERNS
 			uses reusing %2, REG={OR_RIS, %2, hi(%1.val)}
 			yields {OR_RC, %2, lo(%1.val)}
 
-	pat ior !defined($1)               /* OR set */
+	pat ior defined($1)                /* OR set */
 		with STACK
+			kills ALL
+			gen
+				move {CONST, $1}, R3
+				bl {LABEL, ".ior"}
+
+	/* OR set (variable), used in lang/m2/libm2/LtoUset.e */
+	pat ior !defined($1)
+		with GPR3 STACK
+			kills ALL
 			gen
 				bl {LABEL, ".ior"}
 
@@ -1523,9 +1534,11 @@ PATTERNS
 			uses reusing %2, REG={XOR_RIS, %2, hi(%1.val)}
 			yields {XOR_RC, %2, lo(%1.val)}
 
-	pat xor !defined($1)               /* XOR set */
+	pat xor defined($1)                /* XOR set */
 		with STACK
+			kills ALL
 			gen
+				move {CONST, $1}, R3
 				bl {LABEL, ".xor"}
 
 	pat com $1==INT32                  /* NOT word */
@@ -1547,11 +1560,23 @@ PATTERNS
 		with GPR
 			yields {NOT_R, %1}
 
-	pat com !defined($1)               /* NOT set */
+	pat com defined($1)                /* NOT set */
 		with STACK
 			gen
+				move {CONST, $1}, R3
 				bl {LABEL, ".com"}
 
+	pat zer $1==4                      /* Push zero */
+		leaving
+			loc 0
+
+	pat zer defined($1)	   	   /* Create empty set */
+		with STACK
+			kills ALL
+			gen
+				move {CONST, $1}, R3
+				bl {LABEL, ".zer"}
+
 	pat sli $1==4                      /* Shift left (second << top) */
 		with CONST_ALL GPR
 			uses reusing %2, REG
@@ -1594,6 +1619,7 @@ PATTERNS
 
 	pat aar $1==INT32                  /* Index array */
 		with GPR3 GPR4 GPR5
+			kills ALL
 			gen
 				bl {LABEL, ".aar4"}
 			yields R3
@@ -1624,21 +1650,23 @@ PATTERNS
 
 
 
-			
 /* Sets */
 
-	pat set defined($1)                /* Create word with set bit */
+	pat set defined($1)                /* Create singleton set */
-		leaving
+		with GPR4 STACK
-			loc 1
+			kills ALL
-			exg INT32
+			gen
-			sli INT32
+				move {CONST, $1}, R3
+				bl {LABEL, ".set"}
 
-	pat set !defined($1)               /* Create structure with set bit (variable) */
+	/* Create set (variable), used in lang/m2/libm2/LtoUset.e */
+	pat set !defined($1)
 		with GPR3 GPR4 STACK
+			kills ALL
 			gen
 				bl {LABEL, ".set"}
 
-	pat inn                            /* Test for set bit */
+	pat inn defined($1)                /* Test for set bit */
 		with STACK
 			kills ALL
 			uses REG
@@ -1648,7 +1676,6 @@ PATTERNS
 				bl {LABEL, ".inn"}
 
 
-			
 /* Boolean resolutions */
 
 	pat teq                            /* top = (top == 0) */
@@ -1785,13 +1812,13 @@ PATTERNS
 
 /* Compare and jump */
 
-	pat cmi                            /* Signed tristate compare */
+	pat cmi $1==INT32                  /* Signed tristate compare */
 		with CONST_ALL GPR
 			yields {TRISTATE_RC_S, %2, %1.val}
 		with GPR GPR
 			yields {TRISTATE_RR_S, %2, %1}
 
-	pat cmu                            /* Unsigned tristate compare */
+	pat cmu $1==INT32                  /* Unsigned tristate compare */
 		with CONST_ALL GPR
 			yields {TRISTATE_RC_U, %2, %1.val}
 		with GPR GPR
@@ -1805,6 +1832,13 @@ PATTERNS
 		leaving
 			cmi INT32
 
+	pat cms defined($1)
+		with STACK
+			kills ALL
+			gen
+				move {CONST, $1}, R3
+				bl {LABEL, ".cms"}
+			yields R3
 
 
 
@@ -1993,10 +2027,6 @@ PATTERNS
 			move SP, %a
 		yields %a
 
-	pat lor $1==2                      /* Load HP */
-		leaving
-			loe ".reghp"
-			
 	pat str $1==0                      /* Store FP */
 		with GPR
 			gen
@@ -2007,18 +2037,14 @@ PATTERNS
 			gen
 				move %1, SP
 
-	pat str $1==2                      /* Store HP */
+	pat loc ass $1==4 && $2==4         /* Drop 4 bytes from stack */
-		leaving
-			ste ".reghp"
-
-	pat loc ass $1==4                  /* Drop 4 bytes from stack */
 		with exact GPR
 			/* nop */
 		with STACK
 			gen
 				addi SP, SP, {CONST, 4}
 
-	pat ass                            /* Adjust stack by variable amount */
+	pat ass $1==4                      /* Adjust stack by variable amount */
 		with CONST2 STACK
 			gen
 				move {SUM_RC, SP, %1.val}, SP
@@ -2036,7 +2062,7 @@ PATTERNS
 	pat asp                            /* Adjust stack by constant amount */
 		leaving
 			loc $1
-			ass
+			ass 4