POWER10: Improve axpy performance

This patch aligns the stores to 32 byte boundary for saxpy and daxpy
before entering into vector pair loop. Fox caxpy, changed the store
instructions to stxv to improve performance of unaligned cases.

kernel/power/caxpy_microk_power10.c

@@ -112,10 +112,14 @@ static void caxpy_kernel_8 (long n, float *x, float *y,
        "xvmaddasp	38, 58, 33	\n\t"
        "xvmaddasp	39, 59, 33	\n\t"
 
-       "stxvp		48, 0(%4)	\n\t"
-       "stxvp		50, 32(%4)	\n\t"
-       "stxvp		34, 64(%4)	\n\t"
-       "stxvp		38, 96(%4)	\n\t"
+       "stxv		49, 0(%4)	\n\t"
+       "stxv		48, 16(%4)	\n\t"
+       "stxv		51, 32(%4)	\n\t"
+       "stxv		50, 48(%4)	\n\t"
+       "stxv		35, 64(%4)	\n\t"
+       "stxv		34, 80(%4)	\n\t"
+       "stxv		39, 96(%4)	\n\t"
+       "stxv		38, 112(%4)	\n\t"
 
        "addi		%4, %4, 128	\n\t"
        "xxperm 52, 40, %x10 \n\t"       // exchange real and imag part
@@ -159,10 +163,14 @@ static void caxpy_kernel_8 (long n, float *x, float *y,
        "xvmaddasp	38, 58, 33	\n\t"
        "xvmaddasp	39, 59, 33	\n\t"
 
-       "stxvp		48, 0(%4)	\n\t"
-       "stxvp		50, 32(%4)	\n\t"
-       "stxvp		34, 64(%4)	\n\t"
-       "stxvp		38, 96(%4)	\n\t"
+       "stxv		49, 0(%4)	\n\t"
+       "stxv		48, 16(%4)	\n\t"
+       "stxv		51, 32(%4)	\n\t"
+       "stxv		50, 48(%4)	\n\t"
+       "stxv		35, 64(%4)	\n\t"
+       "stxv		34, 80(%4)	\n\t"
+       "stxv		39, 96(%4)	\n\t"
+       "stxv		38, 112(%4)	\n\t"
 
      "#n=%1 x=%5=%2 y=%0=%3 alpha=(%7,%8) mvecp=%6=%9 ytmp=%4\n"
      :

kernel/power/daxpy_power10.c

@@ -66,12 +66,19 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 	if ( (inc_x == 1) && (inc_y == 1) )
 	{
 
-		BLASLONG n1 = n & -16;
+                if ( n >= 16 )
+                {
+                       BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 3) & 0x3;
+                        for (i = 0; i < align; i++) {
+                          y[i] += da * x[i] ;
+                        }
+                }
+                BLASLONG n1 = (n-i) & -16;
+                if ( n1 )
+                      daxpy_kernel_8(n1, &x[i], &y[i], da);
+
+                i += n1;
 
-		if ( n1 )
-			daxpy_kernel_8(n1, x, y, da);
-
-		i = n1;
 		while(i < n)
 		{
 

kernel/power/saxpy_power10.c

@@ -64,12 +64,18 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
 	if ( (inc_x == 1) && (inc_y == 1) )
 	{
 
-		BLASLONG n1 = n & -64;
-
+		if ( n >= 64 )
+		{
+			BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 2) & 0x7;
+			for (i = 0; i < align; i++) {
+				y[i] += da * x[i] ;
+			}
+		}
+		BLASLONG n1 = (n-i) & -64;
 		if ( n1 )
-			saxpy_kernel_64(n1, x, y, da);
+			saxpy_kernel_64(n1, &x[i], &y[i], da);
 
-		i = n1;
+		i += n1;
 		while(i < n)
 		{