Merge pull request #5448 from martin-frbg/issue5372-2

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

Author: martin-frbg

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