Merge pull request #5282 from martin-frbg/zscal_power

kernel/power: Fixed cscal and zscal

kernel/power/zscal.S

@@ -51,6 +51,7 @@
 #define X r8
 #define INCX r9
 #endif
+#define FLAG r11
 #endif
 
 #if defined(_AIX) || defined(__APPLE__)
@@ -61,6 +62,7 @@
 #define X r8
 #define INCX r9
 #endif
+#define FLAG r11
 #endif
 
 #define FZERO	f0
@@ -94,6 +96,10 @@
 	fcmpu	cr0, FZERO, ALPHA_I
 	bne-	cr0, LL(A1I1)
 
+	LDLONG	FLAG, 104(SP)
+	cmpwi	cr0, FLAG, 1
+	beq-	cr0, LL(A1I1)
+
 	cmpwi	cr0, INCX, 2 * SIZE
 	bne-	cr0, LL(A0IN)
 

kernel/power/zscal.c

@@ -136,7 +136,7 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da_r,FLOAT da_i, F
 	if ( inc_x <= 0 )
 		return(0);
 
-	if (da_r == ZERO && da_i == ZERO) {
+	if (da_r == ZERO && da_i == ZERO && dummy2 == 0) {
 	  //clear the vector and return
 	  if (inc_x == 1) {
 	    memset(x, 0, n*COMPSIZE*SIZE);

kernel/power/zscal_ppc440.S

@@ -64,6 +64,7 @@
 #endif
 
 #define INC1	r11
+#define FLAG	r12
 
 #define FZERO	f0
 #define ALPHA_R	f1
@@ -97,6 +98,10 @@
 	fcmpu	cr0, FZERO, ALPHA_I
 	bne-	cr0, LL(A1I1)
 
+	lwz	FLAG, FRAMESLOT(0)(SP)
+	cmpwi	cr0, FLAG, 1
+	beq-	cr0, LL(A1I1)
+
 LL(A0IN):
 	srawi.	r0, N, 3
 	mtspr	CTR,  r0