Fix clobber list entries for arrays in POWER kernels that use inline asm

kernel/power/casum_microk_power8.c

@@ -159,7 +159,7 @@ static float casum_kernel_16 (long n, float *x)
        "=wa" (t2),	// 5
        "=wa" (t3)	// 6
      :
-       "m" (*x),
+       "m" (*(const float (*)[n * 2]) x),
        "b" (16),	// 8
        "b" (32),	// 9
        "b" (48),	// 10

kernel/power/caxpy_microk_power10.c

@@ -199,14 +199,14 @@ static void caxpy_kernel_8 (long n, float *x, float *y,
 
      "#n=%1 x=%5=%2 y=%0=%3 alpha=(%7,%8) mvecp=%6=%9 ytmp=%4\n"
      :
-       "+m" (*y),
+       "+m" (*(float (*)[n * 2]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y),	// 3
        "=b" (ytmp)	// 4 
      :
-       "m" (*x),
-       "m" (*mvecp),
+       "m" (*(const float (*)[n * 2]) x),
+       "m" (*(const float (*)[2]) mvecp),
        "d" (alpha_r),	// 7
        "d" (alpha_i),	// 8
        "4" (mvecp),	// 9

kernel/power/ccopy_microk_power8.c

@@ -134,12 +134,12 @@ static void ccopy_kernel_32 (long n, float *x, float *y)
 
      "#n=%1 x=%4=%2 y=%0=%3 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :
-       "=m" (*y),
+       "=m" (*(float (*)[n * 2]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
+       "m" (*(const float (*)[n * 2]) x),
        "b" (16),	// 5
        "b" (32),	// 6
        "b" (48),	// 7

kernel/power/cdot_microk_power10.c

@@ -166,13 +166,13 @@ static void cdot_kernel_8 (long n, float *x, float *y, float *dot)
 
      "#n=%1 x=%4=%2 y=%5=%3 dot=%0=%6"
      :
-       "=m" (*dot),
+       "=m" (*(float (*)[4]) dot),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
-       "m" (*y),
+       "m" (*(float (*)[n * 2]) x),
+       "m" (*(float (*)[n * 2]) y),
        "b" (dot),	// 6
        "wa" (mask)
      :

kernel/power/copy_microk_power10.c

@@ -241,12 +241,12 @@ static void copy_kernel (BLASLONG n, FLOAT *x, FLOAT *y)
 
      "#n=%1 x=%4=%2 y=%0=%3"
      :
-       "=m" (*y),
+       "=m" (*(FLOAT (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y) 	// 3
      :
-       "m" (*x)
+       "m" (*(FLOAT (*)[n]) x)
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",

kernel/power/cscal_microk_power10.c

@@ -204,7 +204,7 @@ static void zscal_kernel_8 (long n, float *x, float alpha_r, float alpha_i)
 
      "#n=%1 x=%0=%2 alpha=(%3,%4)\n"
      :
-       "+m" (*x),
+       "+m" (*(float (*)[n * 2]) x),
        "+r" (n),	// 1
        "+b" (x)		// 2
      :

kernel/power/cswap_microk_power10.c

@@ -155,8 +155,8 @@ static void cswap_kernel_32 (long n, float *x, float *y)
 
      "#n=%2 x=%0=%3 y=%1=%4"
      :
-       "+m" (*x),
-       "+m" (*y),
+       "+m" (*(FLOAT (*)[n * 2]) x),
+       "+m" (*(FLOAT (*)[n * 2]) y),
        "+r" (n),	// 2
        "+b" (x),	// 3
        "+b" (y)		// 4

kernel/power/cswap_microk_power8.c

@@ -135,8 +135,8 @@ static void cswap_kernel_32 (long n, float *x, float *y)
 
      "#n=%2 x=%0=%3 y=%1=%4 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :
-       "+m" (*x),
-       "+m" (*y),
+       "+m" (*(float (*)[n * 2]) x),
+       "+m" (*(float (*)[n * 2]) y),
        "+r" (n),	// 2
        "+b" (x),	// 3
        "+b" (y)		// 4

kernel/power/dasum_microk_power10.c

@@ -225,7 +225,7 @@ static double dasum_kernel_16 (long n, double *x)
        "=wa" (a6),	// 17
        "=wa" (a7)	// 18
      :
-       "m" (*x)
+       "m" (*(const double (*)[n]) x)
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",

kernel/power/dasum_microk_power8.c

@@ -154,7 +154,7 @@ static double dasum_kernel_16 (long n, double *x)
        "=wa" (t2),	// 5
        "=wa" (t3)	// 6
      :
-       "m" (*x),
+       "m" (*(const double (*)[n]) x),
        "b" (16),	// 8
        "b" (32),	// 9
        "b" (48),	// 10

kernel/power/daxpy_microk_power10.c

@@ -150,13 +150,13 @@ static void daxpy_kernel_8 (long n, double *x, double *y, double alpha)
 
      "#n=%1 x=%5=%2 y=%0=%3 alpha=%6 t0=%x4\n"
      :
-       "+m" (*y),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y),	// 3
        "=wa" (t0)	// 4
      :
-       "m" (*x),
+       "m" (*(const double (*)[n]) x),
        "d" (alpha)	// 6 
      :
        "cr0",

kernel/power/daxpy_microk_power8.c

@@ -181,7 +181,7 @@ static void daxpy_kernel_8 (long n, double *x, double *y, double alpha)
      "#n=%1 x=%21=%2 y=%0=%3 alpha=%22 o16=%23 o32=%24 o48=%25\n"
      "#t0=%x4 t1=%x5 t2=%x6 t3=%x7 t4=%x8 t5=%x9 t6=%x10 t7=%x11 t8=%x12 t9=%x13 t10=%x14 t11=%x15 t12=%x16 t13=%x17 t14=%x18 t15=%x19 t16=%x20"
      :
-       "+m" (*y),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y),	// 3
@@ -203,7 +203,7 @@ static void daxpy_kernel_8 (long n, double *x, double *y, double alpha)
        "=wa" (t15),	// 19
        "=wa" (t16)	// 20
      :
-       "m" (*x),
+       "m" (*(const double (*)[n]) x),
        "d" (alpha),	// 22
        "b" (16),	// 23
        "b" (32),	// 24

kernel/power/dcopy_microk_power8.c

@@ -134,12 +134,12 @@ static void dcopy_kernel_32 (long n, double *x, double *y)
 
      "#n=%1 x=%4=%2 y=%0=%3 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :
-       "=m" (*y),
+       "=m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
+       "m" (*(const double (*)[n]) x),
        "b" (16),	// 5
        "b" (32),	// 6
        "b" (48),	// 7

kernel/power/ddot_microk_power10.c

@@ -118,8 +118,8 @@ static double ddot_kernel_8 (long n, double *x, double *y)
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
-       "m" (*y)
+       "m" (*(const double (*)[n]) x),
+       "m" (*(const double (*)[n]) y),
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",

kernel/power/ddot_microk_power8.c

@@ -151,8 +151,8 @@ static double ddot_kernel_8 (long n, double *x, double *y)
        "=wa" (t2),	// 6
        "=wa" (t3)	// 7
      :
-       "m" (*x),
-       "m" (*y),
+       "m" (*(const double (*)[n]) x),
+       "m" (*(const double (*)[n]) y),
        "b" (16),	// 10
        "b" (32),	// 11
        "b" (48),	// 12

kernel/power/dgemv_n_microk_power10.c

@@ -246,16 +246,16 @@ static void dgemv_kernel_4x4 (long n, double *ap, long lda, double *x, double *y
      "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n"
      "#a0=%3 a1=%4 a2=%5 a3=%6"
      :
-       "+m" (*y),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (y),	// 2
        "=b" (a0),	// 3
        "=b" (a1),	// 4
        "=&b" (a2),	// 5
        "=&b" (a3)	// 6
      :
-       "m" (*x),
-       "m" (*ap),
+       "m" (*(double (*)[4]) x),
+       "m" (*(double (*)[4*n]) ap),
        "d" (alpha),	// 9
        "r" (x),		// 10
        "b" (16),	// 11
@@ -452,7 +452,7 @@ static void dgemv_kernel_4x8 (long n, double *ap, long lda, double *x, double *y
        "stxvp		36, 0( %2)	\n\t"	// y0, y1
 
      :
-       "+m" (*y),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (y),	// 2
        "=b" (a0),	// 3
@@ -465,8 +465,8 @@ static void dgemv_kernel_4x8 (long n, double *ap, long lda, double *x, double *y
        "=&b" (a7),	// 10
        "=b" (tmp)
      :
-       "m" (*x),
-       "m" (*ap),
+       "m" (*(double (*)[4]) x),
+       "m" (*(double (*)[]) ap),
        "d" (alpha),	// 14
        "r" (x),		// 15
        "3" (ap),	// 16

kernel/power/dgemv_n_microk_power8.c

@@ -276,16 +276,16 @@ static void dgemv_kernel_4x4 (long n, double *ap, long lda, double *x, double *y
      "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n"
      "#a0=%3 a1=%4 a2=%5 a3=%6"
      :
-       "+m" (*y),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 1
        "+b" (y),	// 2
        "=b" (a0),	// 3
        "=b" (a1),	// 4
        "=&b" (a2),	// 5
        "=&b" (a3)	// 6
      :
-       "m" (*x),
-       "m" (*ap),
+       "m" (*(double (*)[4]) x),
+       "m" (*(double (*)[4*n]) ap),
        "d" (alpha),	// 9
        "r" (x),		// 10
        "b" (16),	// 11

kernel/power/drot_microk_power10.c

@@ -130,8 +130,8 @@ static void drot_kernel_16 (long n, double *x, double *y, double c, double s)
 
      "#n=%2 x=%0=%3 y=%1=%4 c=%5 s=%6\n"
      :
-       "+m" (*x),
-       "+m" (*y),
+       "+m" (*(double (*)[n]) x),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 2
        "+b" (x),	// 3
        "+b" (y) 	// 4

kernel/power/drot_microk_power8.c

@@ -185,8 +185,8 @@ static void drot_kernel_16 (long n, double *x, double *y, double c, double s)
      "#n=%2 x=%0=%3 y=%1=%4 c=%13 s=%14 o16=%15 o32=%16 o48=%17\n"
      "#t0=%x5 t1=%x6 t2=%x7 t3=%x8 t4=%x9 t5=%x10 t6=%x11 t7=%x12"
      :
-       "+m" (*x),
-       "+m" (*y),
+       "+m" (*(double (*)[n]) x),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 2
        "+b" (x),	// 3
        "+b" (y),	// 4

kernel/power/dscal_microk_power10.c

@@ -118,7 +118,7 @@ static void dscal_kernel_8 (long n, double *x, double alpha)
 
      "#n=%1 alpha=%3 x=%0=%2"
      :
-       "+m" (*x),
+       "+m" (*(double (*)[n]) x),
        "+r" (n),	// 1
        "+b" (x)		// 2
      :
@@ -158,7 +158,7 @@ static void dscal_kernel_8_zero (long n, double *x)
 
      "#n=%1 x=%0=%2 "
      :
-       "=m" (*x),
+       "=m" (*(double (*)[n]) x),
        "+r" (n),	// 1
        "+b" (x)	// 2
      :

kernel/power/dscal_microk_power8.c

@@ -118,7 +118,7 @@ static void dscal_kernel_8 (long n, double *x, double alpha)
 
      "#n=%1 alpha=%3 x=%0=%2 o16=%4 o32=%5 o48=%6 o64=%7 o80=%8 o96=%9 o112=%10"
      :
-       "+m" (*x),
+       "+m" (*(double (*)[n]) x),
        "+r" (n),	// 1
        "+b" (x)		// 2
      :
@@ -165,7 +165,7 @@ static void dscal_kernel_8_zero (long n, double *x)
 
      "#n=%1 x=%0=%2 t0=%x3 o16=%4 o32=%5 o48=%6 o64=%7 o80=%8 o96=%9 o112=%10"
      :
-       "=m" (*x),
+       "=m" (*(double (*)[n]) x),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "=wa" (t0)	// 3

kernel/power/dswap_microk_power8.c

@@ -135,8 +135,8 @@ static void dswap_kernel_32 (long n, double *x, double *y)
 
      "#n=%2 x=%0=%3 y=%1=%4 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :
-       "+m" (*x),
-       "+m" (*y),
+       "+m" (*(double (*)[n]) x),
+       "+m" (*(double (*)[n]) y),
        "+r" (n),	// 2
        "+b" (x),	// 3
        "+b" (y)		// 4

kernel/power/sasum_microk_power10.c

@@ -141,7 +141,7 @@ static float sasum_kernel_32 (long n, float *x)
        "=wa" (t2),	// 5
        "=wa" (t3)	// 6
      :
-       "m" (*x)
+       "m" (*(const float (*)[n]) x),
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",

kernel/power/sasum_microk_power8.c

@@ -159,7 +159,7 @@ static float sasum_kernel_32 (long n, float *x)
        "=wa" (t2),	// 5
        "=wa" (t3)	// 6
      :
-       "m" (*x),
+       "m" (*(const float (*)[n]) x),
        "b" (16),	// 8
        "b" (32),	// 9
        "b" (48),	// 10

kernel/power/saxpy_microk_power10.c

@@ -238,13 +238,13 @@ static void saxpy_kernel_64(long n, float *x, float *y, float alpha)
 
      "#n=%1 x=%5=%2 y=%0=%3 t0=%x4\n"
      :
-       "+m" (*y),
+       "+m" (*(float (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
        "wa" (t0),	// 4
-       "m" (*x)
+       "m" (*(const float (*)[n]) x)
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37", "vs38", "vs39",

kernel/power/scopy_microk_power8.c

@@ -92,12 +92,12 @@ static void scopy_kernel_32 (long n, float *x, float *y)
 
      "#n=%1 x=%4=%2 y=%0=%3 o16=%5 o32=%6 o48=%7 o64=%8 o80=%9 o96=%10 o112=%11"
      :
-       "=m" (*y),
+       "=m" (*(float (*)[n]) y),
        "+r" (n),	// 1
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
+       "m" (*(const float (*)[n]) x),
        "b" (16),	// 5
        "b" (32),	// 6
        "b" (48),	// 7

kernel/power/sdot_microk_power10.c

@@ -122,8 +122,8 @@ static float sdot_kernel_16 (long n, float *x, float *y)
        "+b" (x),	// 2
        "+b" (y)		// 3
      :
-       "m" (*x),
-       "m" (*y)
+       "m" (*(const float (*)[n]) x),
+       "m" (*(const float (*)[n]) y)
      :
        "cr0",
        "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",

kernel/power/sdot_microk_power8.c

@@ -155,8 +155,8 @@ static float sdot_kernel_16 (long n, float *x, float *y)
        "=wa" (t2),	// 6
        "=wa" (t3)	// 7
      :
-       "m" (*x),
-       "m" (*y),
+       "m" (*(const float (*)[n]) x),
+       "m" (*(const float (*)[n]) y),
        "b" (16),	// 10
        "b" (32),	// 11
        "b" (48),	// 12