[ZARCH] Fix constraints and source code formatting

kernel/zarch/camax.c

@@ -136,7 +136,7 @@ static FLOAT camax_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmaxsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/camin.c

@@ -136,7 +136,7 @@ static FLOAT camin_kernel_32(BLASLONG n, FLOAT *x) {
     "wfminsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/casum.c

@@ -107,8 +107,8 @@ static FLOAT casum_kernel_32(BLASLONG n, FLOAT *x) {
     "vrepf   %%v25,%%v24,2\n\t"
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[asum],0"
-       : [asum] "=m"(asum),[n] "+&r"(n)
+    : [asum] "=Q"(asum),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/caxpy.c

@@ -91,9 +91,9 @@ static void caxpy_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
     "vst %%v19,112(%%r1,%[y])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v8", "v9", "v10", "v11", "v12", "v13",
        "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");

kernel/zarch/ccopy.c

@@ -36,8 +36,9 @@ static void ccopy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "la  %[x],256(%[x])\n\t"
     "la  %[y],256(%[y])\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n * 2]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n * 2]; } *) y),[x] "+&a"(x),[y] "+&a"(y),
-       : "m"(*(const FLOAT (*)[n * 2]) x)
+       [n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x)
     : "cc");
 }
 

kernel/zarch/cdot.c

@@ -97,9 +97,9 @@ static void cdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d) {
     "vstef  %%v24,4(%[d]),1\n\t"
     "vstef  %%v25,8(%[d]),1\n\t"
     "vstef  %%v25,12(%[d]),0"
-       : "=m"(*(FLOAT (*)[4]) d),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[4]; } *) d),[n] "+&r"(n)
-       : [d] "a"(d), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+    : [d] "a"(d), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[n * 2]) y),[y] "a"(y)
+       "m"(*(const struct { FLOAT x[n * 2]; } *) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/cgemv_n_4.c

@@ -30,6 +30,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define NBMAX 2048
 
 static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vlrepg     %%v16,0(%[x])\n\t"
     "vlrepg     %%v17,8(%[x])\n\t"
     "vlrepg     %%v18,16(%[x])\n\t"
@@ -141,18 +146,21 @@ static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vst %%v0,0(%%r1,%[y])\n\t"
     "agfi   %%r1,16\n\t"
     "brctg  %[n],0b\n\t"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n * 2]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[8]) x),[x] "a"(x)
+       "m"(*(const struct { FLOAT x[8]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
 }
 
 static void cgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vlrepg     %%v16,0(%[x])\n\t"
     "vlrepg     %%v17,8(%[x])\n\t"
 #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
@@ -230,10 +238,10 @@ static void cgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vst %%v0,0(%%r1,%[y])\n\t"
     "agfi   %%r1,16\n\t"
     "brctg  %[n],0b\n\t"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[4]) x),[x] "a"(x)
+       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23");
 }
@@ -299,9 +307,9 @@ static void cgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y) {
     "vst %%v0,0(%%r1,%[y])\n\t"
     "agfi   %%r1,16\n\t"
     "brctg  %[n],0b\n\t"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-          "m"(*(const FLOAT (*)[2]) x),[x] "a"(x)
+       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19");
 }
 
@@ -342,9 +350,9 @@ static void add_y_4(BLASLONG n, FLOAT *src, FLOAT *dest, FLOAT alpha_r,
     "vst %%v23,16(%%r1,%[dest])\n\t"
     "agfi   %%r1,32\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) dest),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) dest),[n] "+&r"(n)
-       : [dest] "a"(dest), "m"(*(const FLOAT (*)[n * 2]) src),[src] "a"(src),
+    : [dest] "a"(dest), "m"(*(const struct { FLOAT x[n * 2]; } *) src),
-          [alpha_r] "m"(alpha_r),[alpha_i] "m"(alpha_i)
+       [src] "a"(src),[alpha_r] "m"(alpha_r),[alpha_i] "m"(alpha_i)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23");
 }

kernel/zarch/cgemv_t_4.c

@@ -31,6 +31,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vzero  %%v16\n\t"
         "vzero  %%v17\n\t"
         "vzero  %%v18\n\t"
@@ -154,13 +159,13 @@ static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
         "vfmasb   %%v23,%%v19,%%v21,%%v23\n\t"
         "vst  %%v22,0(%[y])\n\t"
         "vst  %%v23,16(%[y])"
-       : "+m"(*(FLOAT (*)[8]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[8]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n * 2]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
@@ -168,6 +173,9 @@ static void cgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
 
 static void cgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vzero  %%v16\n\t"
         "vzero  %%v17\n\t"
         "vzero  %%v18\n\t"
@@ -263,11 +271,11 @@ static void cgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
         "vfmasb   %%v20,%%v16,%%v18,%%v20\n\t"
         "vfmasb   %%v20,%%v17,%%v19,%%v20\n\t"
         "vst  %%v20,0(%[y])"
-       : "+m"(*(FLOAT (*)[4]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[4]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23");
 }
@@ -353,10 +361,10 @@ static void cgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y,
         "vfmasb   %%v0,%%v16,%%v18,%%v0\n\t"
         "vfmasb   %%v0,%%v17,%%v19,%%v0\n\t"
         "vsteg    %%v0,0(%[y]),0"
-       : "+m"(*(FLOAT (*)[2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19");
 }
 

kernel/zarch/crot.c

@@ -169,8 +169,9 @@ static void crot_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s) {
     "vst  %%v23, 240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x), "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),
-       : [x] "a"(x),[y] "a"(y),[c] "m"(*c),[s] "m"(*s)
+       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : [x] "a"(x),[y] "a"(y),[c] "Q"(*c),[s] "Q"(*s)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/cscal.c

@@ -80,8 +80,9 @@ static void cscal_kernel_16(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
@@ -131,8 +132,9 @@ static void cscal_kernel_16_zero_r(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
 }
@@ -169,8 +171,9 @@ static void cscal_kernel_16_zero_i(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
 }
@@ -191,7 +194,7 @@ static void cscal_kernel_16_zero(BLASLONG n, FLOAT *x) {
     "vst  %%v0,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/cswap.c

@@ -99,7 +99,8 @@ static void cswap_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "vst  %%v31, 240(%%r1,%[y])\n\t"
     "agfi   %%r1,256\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x), "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),
+       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",

kernel/zarch/damax.c

@@ -76,7 +76,7 @@ static FLOAT damax_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmaxdb %%v0,%%v0,%%v16,8\n\t"
     "lpdr    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/damax_z13.c

@@ -110,7 +110,7 @@ static FLOAT damax_kernel_32(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/damin.c

@@ -76,7 +76,7 @@ static FLOAT damin_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmindb %%v0,%%v0,%%v16,8\n\t"
     "lpdr    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/damin_z13.c

@@ -110,7 +110,7 @@ static FLOAT damin_kernel_32(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/dasum.c

@@ -105,8 +105,8 @@ static FLOAT dasum_kernel_32(BLASLONG n, FLOAT *x) {
     "vrepg   %%v25,%%v24,1\n\t"
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[asum],0"
-       : [asum] "=m"(asum),[n] "+&r"(n)
+    : [asum] "=Q"(asum),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/daxpy.c

@@ -100,9 +100,9 @@ static void daxpy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
     "vst  %%v27,240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
-          [alpha] "m"(*alpha)
+       [alpha] "Q"(*alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 }

kernel/zarch/dcopy.c

@@ -36,8 +36,8 @@ static void dcopy_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "la  %[x],256(%[x])\n\t"
     "la  %[y],256(%[y])\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n]; } *) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x)
     : "cc");
 }
 

kernel/zarch/ddot.c

@@ -80,8 +80,8 @@ static FLOAT ddot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y) {
     "adbr   %%f0,%%f1\n\t"
     "ldr    %[dot],%%f0"
     : [dot] "=f"(dot),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x), "m"(*(const FLOAT (*)[n]) y),
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
-          [y] "a"(y)
+       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/dgemv_n_4.c

@@ -31,6 +31,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vlrepg %%v0,0(%[x])\n\t"
     "vlrepg %%v1,8(%[x])\n\t"
     "vlrepg %%v2,16(%[x])\n\t"
@@ -165,12 +170,13 @@ static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[4]) x),[x] "a"(x),[alpha] "m"(*alpha),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+       [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -178,6 +184,9 @@ static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
 
 static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vlrepg %%v0,0(%[x])\n\t"
     "vlrepg %%v1,8(%[x])\n\t"
     "vlrepg %%v2,%[alpha]\n\t"
@@ -266,10 +275,11 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[2]) x),[x] "a"(x),[alpha] "m"(*alpha),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+       [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v8", "v9", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
@@ -342,9 +352,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-          "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "m"(*alpha),
+       "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "Q"(*alpha),
        [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",

kernel/zarch/dgemv_t_4.c

@@ -30,6 +30,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define NBMAX 2048
 
 static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vzero %%v0\n\t"
     "vzero %%v1\n\t"
     "vzero %%v2\n\t"
@@ -168,18 +173,21 @@ static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vrepg  %%v4,%%v3,1\n\t"
     "adbr   %%f3,%%f4\n\t"
     "std    %%f3,24(%[y])"
-       : "=m"(*(FLOAT (*)[4]) y)
+    : "=m"(*(struct { FLOAT x[4]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
 }
 
 static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vzero %%v0\n\t"
     "vzero %%v1\n\t"
     "vzero %%v2\n\t"
@@ -272,10 +280,10 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vrepg  %%v2,%%v1,1\n\t"
     "adbr   %%f1,%%f2\n\t"
     "std    %%f1,8(%[y])"
-       : "=m"(*(FLOAT (*)[2]) y)
+    : "=m"(*(struct { FLOAT x[2]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -352,8 +360,8 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y) {
     "adbr   %%f0,%%f1\n\t"
     "std    %%f0,0(%[y])"
     : "=m"(*(FLOAT (*)[1]) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -431,8 +439,8 @@ static void add_y_kernel_4(BLASLONG n, FLOAT da, FLOAT *src, FLOAT *dest) {
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) dest)
+    : "+m"(*(struct { FLOAT x[n]; } *) dest)
-       : [dest] "a"(dest),[da] "m"(da), "m"(*(const FLOAT (*)[n]) src),
+    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const struct { FLOAT x[n]; } *) src),
        [src] "a"(src),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",

kernel/zarch/dmax.c

@@ -73,7 +73,7 @@ static FLOAT dmax_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmaxdb %%v0,%%v0,%%v16,0\n\t"
     "ldr    %[max],%%f0"
     : [max] "=f"(max),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/dmax_z13.c

@@ -90,7 +90,7 @@ static FLOAT dmax_kernel_32(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[max],%%f0"
     : [max] "=f"(max),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/dmin.c

@@ -73,7 +73,7 @@ static FLOAT dmin_kernel_32(BLASLONG n, FLOAT *x) {
     "wfmindb %%v0,%%v0,%%v16,0\n\t"
     "ldr    %[min],%%f0"
     : [min] "=f"(min),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/dmin_z13.c

@@ -90,7 +90,7 @@ static FLOAT dmin_kernel_32(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[min],%%f0"
     : [min] "=f"(min),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/drot.c

@@ -169,8 +169,9 @@ static void drot_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s) {
     "vst  %%v23, 240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
-       : [x] "a"(x),[y] "a"(y),[c] "m"(*c),[s] "m"(*s)
+       [n] "+&r"(n)
+    : [x] "a"(x),[y] "a"(y),[c] "Q"(*c),[s] "Q"(*s)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/dscal.c

@@ -59,8 +59,8 @@ static void dscal_kernel_16(BLASLONG n, FLOAT da, FLOAT *x) {
     "vst   %%v31,112(%%r1,%[x])\n\t"
     "agfi   %%r1,128\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x),[da] "m"(da)
+    : [x] "a"(x),[da] "Q"(da)
     : "cc", "r1", "v0", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
 }
@@ -81,7 +81,7 @@ static void dscal_kernel_16_zero(BLASLONG n, FLOAT *x) {
     "vst  %%v0,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/dsdot.c

@@ -112,7 +112,8 @@ static double dsdot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y) {
     "adbr   %%f0,%%f1\n\t"
     "ldr    %[dot],%%f0"
     : [dot] "=f"(dot),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x), "m"(*(const FLOAT (*)[n]) y),[y] "a"(y)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/dswap.c

@@ -99,7 +99,8 @@ static void dswap_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "vst  %%v31, 240(%%r1,%[y])\n\t"
     "agfi   %%r1,256\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+       [n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",

kernel/zarch/icamax.c

@@ -214,8 +214,8 @@ static BLASLONG icamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamax] "=r"(iamax),[amax] "=m"(*amax),[n] "+&r"(n)
+    : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8",
        "v9", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24",
        "v25", "v26", "v27", "v28", "v29", "v30", "v31");

kernel/zarch/icamin.c

@@ -214,8 +214,8 @@ static BLASLONG icamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamin] "=r"(iamin),[amin] "=m"(*amin),[n] "+&r"(n)
+    : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8",
        "v9", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24",
        "v25", "v26", "v27", "v28", "v29", "v30", "v31");

kernel/zarch/idamax.c

@@ -161,8 +161,8 @@ static BLASLONG idamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamax] "=r"(iamax),[amax] "=m"(*amax),[n] "+&r"(n)
+    : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/idamin.c

@@ -161,8 +161,8 @@ static BLASLONG idamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamin] "=r"(iamin),[amin] "=m"(*amin),[n] "+&r"(n)
+    : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/idmax.c

@@ -141,8 +141,8 @@ static BLASLONG idmax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *max) {
     "vlgvg  %[imax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [imax] "=r"(imax),[max] "=m"(*max),[n] "+&r"(n)
+    : [imax] "=r"(imax),[max] "=Q"(*max),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/idmin.c

@@ -141,8 +141,8 @@ static BLASLONG idmin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *min) {
     "vlgvg  %[imin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [imin] "=r"(imin),[min] "=m"(*min),[n] "+&r"(n)
+    : [imin] "=r"(imin),[min] "=Q"(*min),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/isamax.c

@@ -205,8 +205,8 @@ static BLASLONG isamax_kernel_64(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamax] "=r"(iamax),[amax] "=m"(*amax),[n] "+&r"(n)
+    : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v4", "v5", "v6", "v7", "v8", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/isamin.c

@@ -205,8 +205,8 @@ static BLASLONG isamin_kernel_64(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamin] "=r"(iamin),[amin] "=m"(*amin),[n] "+&r"(n)
+    : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v4", "v5", "v6", "v7", "v8", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/ismax.c

@@ -185,8 +185,8 @@ static BLASLONG ismax_kernel_64(BLASLONG n, FLOAT *x, FLOAT *max) {
     "vlgvg  %[imax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [imax] "=r"(imax),[max] "=m"(*max),[n] "+&r"(n)
+    : [imax] "=r"(imax),[max] "=Q"(*max),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v4", "v5", "v6", "v7", "v8", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/ismin.c

@@ -185,8 +185,8 @@ static BLASLONG ismin_kernel_64(BLASLONG n, FLOAT *x, FLOAT *min) {
     "vlgvg  %[imin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [imin] "=r"(imin),[min] "=m"(*min),[n] "+&r"(n)
+    : [imin] "=r"(imin),[min] "=Q"(*min),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v4", "v5", "v6", "v7", "v8", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/izamax.c

@@ -158,8 +158,8 @@ static BLASLONG izamax_kernel_16(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamax] "=r"(iamax),[amax] "=m"(*amax),[n] "+&r"(n)
+    : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18",
        "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27");
 

kernel/zarch/izamin.c

@@ -158,8 +158,8 @@ static BLASLONG izamin_kernel_16(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
     "nop"
-       : [iamin] "=r"(iamin),[amin] "=m"(*amin),[n] "+&r"(n)
+    : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v16", "v17", "v18",
        "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27");
 

kernel/zarch/samax.c

@@ -78,7 +78,7 @@ static FLOAT samax_kernel_64(BLASLONG n, FLOAT *x) {
     "wfmaxsb %%v0,%%v0,%%v16,8\n\t"
     "lper    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/samin.c

@@ -78,7 +78,7 @@ static FLOAT samin_kernel_64(BLASLONG n, FLOAT *x) {
     "wfminsb %%v0,%%v0,%%v16,8\n\t"
     "lper    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/sasum.c

@@ -107,8 +107,8 @@ static FLOAT sasum_kernel_64(BLASLONG n, FLOAT *x) {
     "vrepf   %%v25,%%v24,2\n\t"
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[asum],0"
-       : [asum] "=m"(asum),[n] "+&r"(n)
+    : [asum] "=Q"(asum),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/saxpy.c

@@ -100,9 +100,9 @@ static void saxpy_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
     "vst  %%v27,240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
-          [alpha] "m"(*alpha)
+       [alpha] "Q"(*alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 }

kernel/zarch/scopy.c

@@ -36,8 +36,8 @@ static void scopy_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y) {
     "la  %[x],256(%[x])\n\t"
     "la  %[y],256(%[y])\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n]; } *) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x)
     : "cc");
 }
 

kernel/zarch/sdot.c

@@ -84,8 +84,8 @@ static FLOAT sdot_kernel_32(BLASLONG n, FLOAT *x, FLOAT *y) {
     "aebr   %%f0,%%f3\n\t"
     "ler    %[dot],%%f0"
     : [dot] "=f"(dot),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x), "m"(*(const FLOAT (*)[n]) y),
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x), 
-          [y] "a"(y)
+       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",
        "v27", "v28", "v29", "v30", "v31");

kernel/zarch/sgemv_n_4.c

@@ -31,6 +31,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vlrepf %%v0,0(%[x])\n\t"
     "vlrepf %%v1,4(%[x])\n\t"
     "vlrepf %%v2,8(%[x])\n\t"
@@ -155,12 +160,13 @@ static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[4]) x),[x] "a"(x),[alpha] "m"(*alpha),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+       [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -168,6 +174,9 @@ static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
 
 static void sgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vlrepf %%v0,0(%[x])\n\t"
     "vlrepf %%v1,4(%[x])\n\t"
     "vlrepf %%v2,%[alpha]\n\t"
@@ -250,10 +259,11 @@ static void sgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[2]) x),[x] "a"(x),[alpha] "m"(*alpha),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+       [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v8", "v9", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
@@ -322,9 +332,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y,
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) y)
+    : "+m"(*(struct { FLOAT x[n]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-          "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "m"(*alpha),
+       "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "Q"(*alpha),
        [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",

kernel/zarch/sgemv_t_4.c

@@ -30,6 +30,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define NBMAX 2048
 
 static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vzero %%v0\n\t"
     "vzero %%v1\n\t"
     "vzero %%v2\n\t"
@@ -167,18 +172,21 @@ static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vrepg  %%v4,%%v3,1\n\t"
     "aebr   %%f3,%%f4\n\t"
     "ste    %%f3,12(%[y])"
-       : "=m"(*(FLOAT (*)[4]) y)
+    : "=m"(*(struct { FLOAT x[4]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
 }
 
 static void sgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vzero %%v0\n\t"
     "vzero %%v1\n\t"
     "vzero %%v2\n\t"
@@ -270,10 +278,10 @@ static void sgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vrepg  %%v2,%%v1,1\n\t"
     "aebr   %%f1,%%f2\n\t"
     "ste    %%f1,4(%[y])"
-       : "=m"(*(FLOAT (*)[2]) y)
+    : "=m"(*(struct { FLOAT x[2]; } *) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -349,8 +357,8 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y) {
     "aebr   %%f0,%%f1\n\t"
     "ste    %%f0,0(%[y])"
     : "=m"(*(FLOAT (*)[1]) y)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-          "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),[n] "r"(n)
+       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
        "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25",
        "v26", "v27", "v28", "v29", "v30", "v31");
@@ -424,8 +432,8 @@ static void add_y_kernel_4(BLASLONG n, FLOAT da, FLOAT *src, FLOAT *dest) {
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
     "nop"
-       : "+m"(*(FLOAT (*)[n]) dest)
+    : "+m"(*(struct { FLOAT x[n]; } *) dest)
-       : [dest] "a"(dest),[da] "m"(da), "m"(*(const FLOAT (*)[n]) src),
+    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const struct { FLOAT x[n]; } *) src),
        [src] "a"(src),[n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",

kernel/zarch/smax.c

@@ -75,7 +75,7 @@ static FLOAT smax_kernel_64(BLASLONG n, FLOAT *x) {
     "wfmaxsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[max],%%f0"
     : [max] "=f"(max),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/smin.c

@@ -75,7 +75,7 @@ static FLOAT smin_kernel_64(BLASLONG n, FLOAT *x) {
     "wfminsb %%v0,%%v0,%%v16,0\n\t"
     "ler    %[min],%%f0"
     : [min] "=f"(min),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/srot.c

@@ -169,8 +169,9 @@ static void srot_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s) {
     "vst  %%v23, 240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
-       : [x] "a"(x),[y] "a"(y),[c] "m"(*c),[s] "m"(*s)
+       [n] "+&r"(n)
+    : [x] "a"(x),[y] "a"(y),[c] "Q"(*c),[s] "Q"(*s)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/sscal.c

@@ -59,8 +59,8 @@ static void sscal_kernel_32(BLASLONG n, FLOAT da, FLOAT *x) {
     "vst   %%v31,112(%%r1,%[x])\n\t"
     "agfi   %%r1,128\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x),[da] "m"(da)
+    : [x] "a"(x),[da] "Q"(da)
     : "cc", "r1", "v0", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
 }
@@ -81,7 +81,7 @@ static void sscal_kernel_32_zero(BLASLONG n, FLOAT *x) {
     "vst  %%v0,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/sswap.c

@@ -99,7 +99,8 @@ static void sswap_kernel_64(BLASLONG n, FLOAT *x, FLOAT *y) {
     "vst  %%v31, 240(%%r1,%[y])\n\t"
     "agfi   %%r1,256\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+       [n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",

kernel/zarch/zamax.c

@@ -114,7 +114,7 @@ static FLOAT zamax_kernel_16(BLASLONG n, FLOAT *x) {
     "wfmaxdb %%v0,%%v0,%%v16,0\n\t"
     "ldr    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/zamax_z13.c

@@ -123,7 +123,7 @@ static FLOAT zamax_kernel_16(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[amax],%%f0"
     : [amax] "=f"(amax),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27");
 

kernel/zarch/zamin.c

@@ -114,7 +114,7 @@ static FLOAT zamin_kernel_16(BLASLONG n, FLOAT *x) {
     "wfmindb %%v0,%%v0,%%v16,0\n\t"
     "ldr    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/zamin_z13.c

@@ -123,7 +123,7 @@ static FLOAT zamin_kernel_16(BLASLONG n, FLOAT *x) {
     "vsel   %%v0,%%v0,%%v16,%%v17\n\t"
     "ldr    %[amin],%%f0"
     : [amin] "=f"(amin),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23", "v24", "v25", "v26", "v27");
 

kernel/zarch/zasum.c

@@ -105,8 +105,8 @@ static FLOAT zasum_kernel_16(BLASLONG n, FLOAT *x) {
     "vrepg   %%v25,%%v24,1\n\t"
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[asum],0"
-       : [asum] "=m"(asum),[n] "+&r"(n)
+    : [asum] "=Q"(asum),[n] "+&r"(n)
-       : "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
 

kernel/zarch/zaxpy.c

@@ -95,9 +95,9 @@ static void zaxpy_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha) {
     "vst %%v19,112(%%r1,%[y])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v8", "v9", "v10", "v11", "v12", "v13",
        "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
        "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");

kernel/zarch/zcopy.c

@@ -36,8 +36,9 @@ static void zcopy_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y) {
     "la  %[x],256(%[x])\n\t"
     "la  %[y],256(%[y])\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n * 2]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n * 2]; } *) y),[x] "+&a"(x),[y] "+&a"(y),
-       : "m"(*(const FLOAT (*)[n * 2]) x)
+       [n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x)
     : "cc");
 }
 

kernel/zarch/zdot.c

@@ -93,9 +93,9 @@ static void zdot_kernel_8(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *d) {
     "vsteg  %%v24,8(%[d]),1\n\t"
     "vsteg  %%v25,16(%[d]),1\n\t"
     "vsteg  %%v25,24(%[d]),0"
-       : "=m"(*(FLOAT (*)[4]) d),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[4]; } *) d),[n] "+&r"(n)
-       : [d] "a"(d), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+    : [d] "a"(d), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[n * 2]) y),[y] "a"(y)
+       "m"(*(const struct { FLOAT x[n * 2]; } *) y),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/zgemv_n_4.c

@@ -30,6 +30,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define NBMAX 1024
 
 static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vl     %%v16,0(%[x])\n\t"
     "vl     %%v17,16(%[x])\n\t"
     "vl     %%v18,32(%[x])\n\t"
@@ -107,18 +112,21 @@ static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vst %%v1,16(%%r1,%[y])\n\t"
     "agfi   %%r1,32\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n * 2]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[8]) x),[x] "a"(x)
+       "m"(*(const struct { FLOAT x[8]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
 }
 
 static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vl     %%v16,0(%[x])\n\t"
     "vl     %%v17,16(%[x])\n\t"
 #if ( !defined(CONJ) && !defined(XCONJ) ) || ( defined(CONJ) && defined(XCONJ) )
@@ -164,10 +172,10 @@ static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y) {
     "vst %%v1,16(%%r1,%[y])\n\t"
     "agfi   %%r1,32\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-         "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+         "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-         "m"(*(const FLOAT (*)[4]) x),[x] "a"(x)
+         "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27");
 }
@@ -202,9 +210,9 @@ static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y) {
     "vst %%v1,16(%%r1,%[y])\n\t"
     "agfi   %%r1,32\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-          "m"(*(const FLOAT (*)[2]) x),[x] "a"(x)
+       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21");
 }
 
@@ -253,9 +261,9 @@ static void add_y_4(BLASLONG n, FLOAT *src, FLOAT *dest, FLOAT alpha_r,
     "vst %%v31,48(%%r1,%[dest])\n\t"
     "agfi   %%r1,64\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) dest),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) dest),[n] "+&r"(n)
-       : [dest] "a"(dest), "m"(*(const FLOAT (*)[n * 2]) src),[src] "a"(src),
+    : [dest] "a"(dest), "m"(*(const struct { FLOAT x[n * 2]; } *) src),
-          [alpha_r] "m"(alpha_r),[alpha_i] "m"(alpha_i)
+       [src] "a"(src),[alpha_r] "m"(alpha_r),[alpha_i] "m"(alpha_i)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/zgemv_t_4.c

@@ -31,6 +31,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+  register FLOAT *ap2 = ap[2];
+  register FLOAT *ap3 = ap[3];
+
   __asm__("vzero  %%v16\n\t"
     "vzero  %%v17\n\t"
     "vzero  %%v18\n\t"
@@ -136,13 +141,13 @@ static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "vst  %%v27,16(%[y])\n\t"
     "vst  %%v28,32(%[y])\n\t"
     "vst  %%v29,48(%[y])"
-       : "+m"(*(FLOAT (*)[8]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[8]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) ap[2]),[ap2] "a"(ap[2]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-          "m"(*(const FLOAT (*)[n * 2]) ap[3]),[ap3] "a"(ap[3]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
@@ -150,6 +155,9 @@ static void zgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
 
 static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
                              FLOAT *alpha) {
+  register FLOAT *ap0 = ap[0];
+  register FLOAT *ap1 = ap[1];
+
   __asm__("vzero  %%v16\n\t"
     "vzero  %%v17\n\t"
     "vzero  %%v18\n\t"
@@ -221,11 +229,11 @@ static void zgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "vfmadb   %%v23,%%v19,%%v21,%%v23\n\t"
     "vst  %%v22,0(%[y])\n\t"
     "vst  %%v23,16(%[y])\n\t"
-       : "+m"(*(FLOAT (*)[4]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[4]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap[0]),[ap0] "a"(ap[0]),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-          "m"(*(const FLOAT (*)[n * 2]) ap[1]),[ap1] "a"(ap[1]),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23");
 }
@@ -286,10 +294,10 @@ static void zgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y,
     "vfmadb   %%v0,%%v16,%%v18,%%v0\n\t"
     "vfmadb   %%v0,%%v17,%%v19,%%v0\n\t"
     "vst  %%v0,0(%[y])\n\t"
-       : "+m"(*(FLOAT (*)[2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[2]; } *) y),[n] "+&r"(n)
-       : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-          "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-          "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19");
 }
 

kernel/zarch/zrot.c

@@ -169,8 +169,9 @@ static void zrot_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *c, FLOAT *s) {
     "vst  %%v23, 240(%%r1,%[y])\n\t"
     "agfi  %%r1,256\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x), "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),
-       : [x] "a"(x),[y] "a"(y),[c] "m"(*c),[s] "m"(*s)
+       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : [x] "a"(x),[y] "a"(y),[c] "Q"(*c),[s] "Q"(*s)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");

kernel/zarch/zscal.c

@@ -78,8 +78,9 @@ static void zscal_kernel_8(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
        "v31");
@@ -127,8 +128,9 @@ static void zscal_kernel_8_zero_r(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
 }
@@ -165,8 +167,9 @@ static void zscal_kernel_8_zero_i(BLASLONG n, FLOAT *alpha, FLOAT *x) {
     "vst %%v23,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-       : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
+    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+       [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
 }
@@ -187,7 +190,7 @@ static void zscal_kernel_8_zero(BLASLONG n, FLOAT *x) {
     "vst  %%v0,112(%%r1,%[x])\n\t"
     "agfi  %%r1,128\n\t"
     "brctg %[n],0b"
-       : "=m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : "=m"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/zswap.c

@@ -99,7 +99,8 @@ static void zswap_kernel_16(BLASLONG n, FLOAT *x, FLOAT *y) {
     "vst  %%v31, 240(%%r1,%[y])\n\t"
     "agfi   %%r1,256\n\t"
     "brctg  %[n],0b"
-       : "+m"(*(FLOAT (*)[n * 2]) x), "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : "+m"(*(struct { FLOAT x[n * 2]; } *) x),
+       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
        "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26",