Merge pull request #80 from xianyi/develop

rebase

Makefile.zarch

@@ -8,3 +8,9 @@ ifeq ($(CORE), Z14)
 CCOMMON_OPT += -march=z14 -mzvector -O3
 FCOMMON_OPT += -march=z14 -mzvector
 endif
+
+# Enable floating-point expression contraction for clang, since it is the
+# default for gcc
+ifeq ($(C_COMPILER), CLANG)
+CCOMMON_OPT += -ffp-contract=fast
+endif

c_check

@@ -8,7 +8,7 @@ $hostos   = `uname -s | sed -e s/\-.*//`;    chop($hostos);
 $hostarch = `uname -m | sed -e s/i.86/x86/`;chop($hostarch);
 $hostarch = `uname -p` if ($hostos eq "AIX");
 $hostarch = "x86_64" if ($hostarch eq "amd64");
-$hostarch = "arm" if ($hostarch =~ /^arm.*/);
+$hostarch = "arm" if ($hostarch ne "arm64" && $hostarch =~ /^arm.*/);
 $hostarch = "arm64" if ($hostarch eq "aarch64");
 $hostarch = "power" if ($hostarch =~ /^(powerpc|ppc).*/);
 $hostarch = "zarch" if ($hostarch eq "s390x");

cmake/system.cmake

@@ -110,6 +110,11 @@ if (NO_AVX2)
   set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX2")
 endif ()
 
+if (NO_AVX512)
+  message(STATUS "Disabling Advanced Vector Extensions 512 (AVX512).")
+  set(GETARCH_FLAGS "${GETARCH_FLAGS} -DNO_AVX512")
+endif ()
+
 if (CMAKE_BUILD_TYPE STREQUAL "Debug")
   set(GETARCH_FLAGS "${GETARCH_FLAGS} ${CMAKE_C_FLAGS_DEBUG}")
 endif ()

cmake/system_check.cmake

@@ -109,6 +109,7 @@ else()
 endif()
 
 if (X86_64 OR X86)
+if (NOT NO_AVX512)
   file(WRITE ${PROJECT_BINARY_DIR}/avx512.c "#include <immintrin.h>\n\nint main(void){ __asm__ volatile(\"vbroadcastss -4 * 4(%rsi), %zmm2\"); }")
 execute_process(COMMAND ${CMAKE_C_COMPILER} -march=skylake-avx512 -c -v -o ${PROJECT_BINARY_DIR}/avx512.o ${PROJECT_BINARY_DIR}/avx512.c OUTPUT_QUIET ERROR_QUIET RESULT_VARIABLE NO_AVX512)
 if (NO_AVX512 EQUAL 1)
@@ -116,6 +117,7 @@ set (CCOMMON_OPT "${CCOMMON_OPT} -DNO_AVX512")
 endif()
   file(REMOVE "avx512.c" "avx512.o")
 endif()
+endif()
 
 include(CheckIncludeFile)
 CHECK_INCLUDE_FILE("stdatomic.h" HAVE_C11)

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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -108,7 +108,7 @@ static FLOAT casum_kernel_32(BLASLONG n, FLOAT *x) {
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[asum],0"
     : [asum] "=Q"(asum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -99,9 +99,9 @@ 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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -36,9 +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"(*(struct { FLOAT x[n * 2]; } *) y),[x] "+&a"(x),[y] "+&a"(y),
+    : "=m"(*(FLOAT (*)[n * 2]) y),[x] "+&a"(x),[y] "+&a"(y),
        [n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x)
+    : "m"(*(const FLOAT (*)[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"(*(struct { FLOAT x[4]; } *) d),[n] "+&r"(n)
-    : [d] "a"(d), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) y),[y] "a"(y)
+    : "=m"(*(FLOAT (*)[4]) d),[n] "+&r"(n)
+    : [d] "a"(d), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -146,12 +146,12 @@ 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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[8]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n * 2]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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");
@@ -238,10 +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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[4]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23");
 }
@@ -307,9 +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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+       "m"(*(const FLOAT (*)[2]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v2", "v16", "v17", "v18", "v19");
 }
 
@@ -350,8 +350,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) dest),[n] "+&r"(n)
-    : [dest] "a"(dest), "m"(*(const struct { FLOAT x[n * 2]; } *) src),
+    : "+m"(*(FLOAT (*)[n * 2]) dest),[n] "+&r"(n)
+    : [dest] "a"(dest), "m"(*(const FLOAT (*)[n * 2]) src),
        [src] "a"(src),[alpha_r] "Q"(alpha_r),[alpha_i] "Q"(alpha_i)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23");

kernel/zarch/cgemv_t_4.c

@@ -159,13 +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"(*(struct { FLOAT x[8]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[8]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n * 2]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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");
@@ -271,11 +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"(*(struct { FLOAT x[4]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[4]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19", "v20",
        "v21", "v22", "v23");
 }
@@ -361,10 +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"(*(struct { FLOAT x[2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19");
 }
 

kernel/zarch/crot.c

@@ -169,8 +169,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),
-       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n * 2]) x),
+       "+m"(*(FLOAT (*)[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",

kernel/zarch/cscal.c

@@ -80,8 +80,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
@@ -132,8 +132,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
@@ -171,8 +171,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
@@ -194,7 +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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
+    : "=m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/csum.c

@@ -90,7 +90,7 @@ static FLOAT csum_kernel_32(BLASLONG n, FLOAT *x) {
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[sum],0"
     : [sum] "=Q"(sum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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/cswap.c

@@ -99,8 +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"(*(struct { FLOAT x[n * 2]; } *) x),
-       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n * 2]) x),
+       "+m"(*(FLOAT (*)[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/ctrmm4x4V.S

@@ -198,7 +198,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,4
       nill LOCAL_VAR1,3 
 #else
-    la LOCAL_VAR1,3(0,0)
+    lghi LOCAL_VAR1,3
     NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x4_BK_Store
@@ -254,7 +254,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x4_BK_Store
@@ -305,7 +305,7 @@ ALIGN_4
        RefreshTempBk LOCAL_VAR1,BK,OFF,1,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/
 #endif 
 jz .L1x4_BK_Store
@@ -385,7 +385,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x2_BK_Store
@@ -442,7 +442,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x2_BK_Store
@@ -492,7 +492,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x2_BK_Store
@@ -568,7 +568,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x1_BK_Store
@@ -620,7 +620,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x1_BK_Store
@@ -670,7 +670,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x1_BK_Store

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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -106,7 +106,7 @@ static FLOAT dasum_kernel_32(BLASLONG n, FLOAT *x) {
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[asum],0"
     : [asum] "=Q"(asum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -100,8 +100,8 @@ 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"(*(struct { FLOAT x[n]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
+    : "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
        [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"(*(struct { FLOAT x[n]; } *) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x)
+    : "=m"(*(FLOAT (*)[n]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
+    : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -169,13 +169,13 @@ static void dgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "agfi   %%r1,32\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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",
@@ -274,11 +274,11 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "agfi   %%r1,32\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[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",
@@ -351,9 +351,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y,
     "agfi   %%r1,32\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
        "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "Q"(*alpha),
        [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",

kernel/zarch/dgemv_t_4.c

@@ -173,12 +173,12 @@ 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"(*(struct { FLOAT x[4]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : "=m"(*(FLOAT (*)[4]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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");
@@ -280,10 +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"(*(struct { FLOAT x[2]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : "=m"(*(FLOAT (*)[2]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[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");
@@ -360,8 +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 struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+       "m"(*(const FLOAT (*)[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");
@@ -438,9 +438,9 @@ static void add_y_kernel_4(BLASLONG n, FLOAT da, FLOAT *src, FLOAT *dest) {
     "agfi   %%r1,32\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) dest)
-    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const struct { FLOAT x[n]; } *) src),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) dest)
+    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -169,7 +169,7 @@ 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"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+    : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) 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",

kernel/zarch/dscal.c

@@ -59,7 +59,7 @@ 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"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
     : [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"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
+    : "=m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/dsdot.c

@@ -112,8 +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 struct { FLOAT x[n]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
+    : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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/dsum.c

@@ -88,7 +88,7 @@ static FLOAT dsum_kernel_32(BLASLONG n, FLOAT *x) {
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[sum],0"
     : [sum] "=Q"(sum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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/dswap.c

@@ -99,7 +99,7 @@ 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"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+    : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),
        [n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",

kernel/zarch/gemm8x4V.S

@@ -147,7 +147,7 @@ brctg LOCAL_VAR1,.L8x4_4_BK
 
 ALIGN_4
 .L8x4_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L8x4_BK_Store
 
@@ -183,7 +183,7 @@ brctg LOCAL_VAR1,.L4x4_4_BK
 
 ALIGN_4
 .L4x4_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L4x4_BK_Store
 
@@ -217,7 +217,7 @@ brctg LOCAL_VAR1,.L2x4_4_BK
 
 ALIGN_4
 .L2x4_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L2x4_BK_Store
 
@@ -252,7 +252,7 @@ brctg LOCAL_VAR1,.L1x4_4_BK
 
 ALIGN_4
 .L1x4_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L1x4_BK_Store
 
@@ -309,7 +309,7 @@ brctg LOCAL_VAR1,.L8x2_4_BK
 
 ALIGN_4
 .L8x2_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L8x2_BK_Store
 
@@ -346,7 +346,7 @@ brctg LOCAL_VAR1,.L4x2_4_BK
 
 ALIGN_4
 .L4x2_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L4x2_BK_Store
 
@@ -380,7 +380,7 @@ brctg LOCAL_VAR1,.L2x2_4_BK
 
 ALIGN_4
 .L2x2_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L2x2_BK_Store
 
@@ -415,7 +415,7 @@ brctg LOCAL_VAR1,.L1x2_4_BK
 
 ALIGN_4
 .L1x2_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L1x2_BK_Store
 
@@ -471,7 +471,7 @@ brctg LOCAL_VAR1,.L8x1_4_BK
 
 ALIGN_4
 .L8x1_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L8x1_BK_Store
 
@@ -508,7 +508,7 @@ brctg LOCAL_VAR1,.L4x1_4_BK
 
 ALIGN_4
 .L4x1_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L4x1_BK_Store
 
@@ -542,7 +542,7 @@ brctg LOCAL_VAR1,.L2x1_4_BK
 
 ALIGN_4
 .L2x1_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L2x1_BK_Store
 
@@ -577,7 +577,7 @@ brctg LOCAL_VAR1,.L1x1_4_BK
 
 ALIGN_4
 .L1x1_mod:
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 jz .L1x1_BK_Store
 

kernel/zarch/gemm_vec.c

@@ -172,7 +172,7 @@ static inline vector_float vec_load_hinted(FLOAT const *restrict a) {
 	vector_float const *restrict addr = (vector_float const *restrict)a;
 	vector_float y;
 
-#if __GNUC__ < 9
+#if __GNUC__ < 9 && !defined(__clang__)
 	// hex-encode vl %[out],%[addr],3
 	asm(".insn vrx,0xe70000003006,%[out],%[addr],3"
 	    : [ out ] "=v"(y)
@@ -393,7 +393,7 @@ static inline void GEBP_block_16_4(
 			 * Note that we need to massage this particular "barrier"
 			 * depending on the gcc version.
 			 */
-#if __GNUC__ > 7
+#if __GNUC__ > 7 || defined(__clang__)
 #define BARRIER_READ_BEFORE_COMPUTE(SUFFIX)                                    \
     do {                                                                       \
 	asm(""                                                                 \

kernel/zarch/icamax.c

@@ -213,9 +213,9 @@ static BLASLONG icamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "ste    %%f0,%[amax]\n\t"
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -213,9 +213,9 @@ static BLASLONG icamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "ste    %%f0,%[amin]\n\t"
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -160,9 +160,9 @@ static BLASLONG idamax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "std    %%f0,%[amax]\n\t"
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -160,9 +160,9 @@ static BLASLONG idamin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "std    %%f0,%[amin]\n\t"
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -140,9 +140,9 @@ static BLASLONG idmax_kernel_32(BLASLONG n, FLOAT *x, FLOAT *max) {
     "std    %%f0,%[max]\n\t"
     "vlgvg  %[imax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [imax] "=r"(imax),[max] "=Q"(*max),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -140,9 +140,9 @@ static BLASLONG idmin_kernel_32(BLASLONG n, FLOAT *x, FLOAT *min) {
     "std    %%f0,%[min]\n\t"
     "vlgvg  %[imin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [imin] "=r"(imin),[min] "=Q"(*min),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -204,9 +204,9 @@ static BLASLONG isamax_kernel_64(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "ste    %%f0,%[amax]\n\t"
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT(*)[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

@@ -204,9 +204,9 @@ static BLASLONG isamin_kernel_64(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "ste    %%f0,%[amin]\n\t"
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -184,9 +184,9 @@ static BLASLONG ismax_kernel_64(BLASLONG n, FLOAT *x, FLOAT *max) {
     "ste    %%f0,%[max]\n\t"
     "vlgvg  %[imax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [imax] "=r"(imax),[max] "=Q"(*max),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -184,9 +184,9 @@ static BLASLONG ismin_kernel_64(BLASLONG n, FLOAT *x, FLOAT *min) {
     "ste    %%f0,%[min]\n\t"
     "vlgvg  %[imin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [imin] "=r"(imin),[min] "=Q"(*min),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -157,9 +157,9 @@ static BLASLONG izamax_kernel_16(BLASLONG n, FLOAT *x, FLOAT *amax) {
     "std    %%f0,%[amax]\n\t"
     "vlgvg  %[iamax],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamax] "=r"(iamax),[amax] "=Q"(*amax),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -157,9 +157,9 @@ static BLASLONG izamin_kernel_16(BLASLONG n, FLOAT *x, FLOAT *amin) {
     "std    %%f0,%[amin]\n\t"
     "vlgvg  %[iamin],%%v1,0\n\t"
     "2:\n\t"
-    "nop"
+    "nop 0"
     : [iamin] "=r"(iamin),[amin] "=Q"(*amin),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -108,7 +108,7 @@ static FLOAT sasum_kernel_64(BLASLONG n, FLOAT *x) {
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[asum],0"
     : [asum] "=Q"(asum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -100,8 +100,8 @@ 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"(*(struct { FLOAT x[n]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),
+    : "+m"(*(FLOAT (*)[n]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
        [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"(*(struct { FLOAT x[n]; } *) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x)
+    : "=m"(*(FLOAT (*)[n]) y),[x] "+&a"(x),[y] "+&a"(y),[n] "+&r"(n)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x), 
-       "m"(*(const struct { FLOAT x[n]; } *) y),[y] "a"(y)
+    : "m"(*(const FLOAT (*)[n]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -159,13 +159,13 @@ static void sgemv_kernel_4x4(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "agfi   %%r1,16\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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",
@@ -258,11 +258,11 @@ static void sgemv_kernel_4x2(BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y,
     "agfi   %%r1,16\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x),[alpha] "Q"(*alpha),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[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",
@@ -331,9 +331,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *a0, FLOAT *x, FLOAT *y,
     "agfi   %%r1,16\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
        "m"(*(const FLOAT (*)[1]) x),[x] "a"(x),[alpha] "Q"(*alpha),
        [n] "r"(n)
     : "cc", "r0", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21",

kernel/zarch/sgemv_t_4.c

@@ -172,12 +172,12 @@ 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"(*(struct { FLOAT x[4]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : "=m"(*(FLOAT (*)[4]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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");
@@ -278,10 +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"(*(struct { FLOAT x[2]; } *) y)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : "=m"(*(FLOAT (*)[2]) y)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[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");
@@ -357,8 +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 struct { FLOAT x[n]; } *) a0),[a0] "a"(a0),
-       "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x),[n] "r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n]) a0),[a0] "a"(a0),
+       "m"(*(const FLOAT (*)[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");
@@ -431,9 +431,9 @@ static void add_y_kernel_4(BLASLONG n, FLOAT da, FLOAT *src, FLOAT *dest) {
     "agfi   %%r1,16\n\t"
     "brctg  %%r0,2b\n\t"
     "3:\n\t"
-    "nop"
-    : "+m"(*(struct { FLOAT x[n]; } *) dest)
-    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const struct { FLOAT x[n]; } *) src),
+    "nop 0"
+    : "+m"(*(FLOAT (*)[n]) dest)
+    : [dest] "a"(dest),[da] "Q"(da), "m"(*(const FLOAT (*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const  FLOAT(*)[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 struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -169,7 +169,7 @@ 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"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+    : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) 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",

kernel/zarch/sscal.c

@@ -59,7 +59,7 @@ 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"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
     : [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"(*(struct { FLOAT x[n]; } *) x),[n] "+&r"(n)
+    : "=m"(*(FLOAT (*)[n]) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/ssum.c

@@ -91,7 +91,7 @@ static FLOAT ssum_kernel_64(BLASLONG n, FLOAT *x) {
     "vfasb   %%v24,%%v24,%%v25\n\t"
     "vstef   %%v24,%[sum],0"
     : [sum] "=Q"(sum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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/sswap.c

@@ -99,7 +99,7 @@ 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"(*(struct { FLOAT x[n]; } *) x), "+m"(*(struct { FLOAT x[n]; } *) y),
+    : "+m"(*(FLOAT (*)[n]) x), "+m"(*(FLOAT (*)[n]) y),
        [n] "+&r"(n)
     : [x] "a"(x),[y] "a"(y)
     : "cc", "r1", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",

kernel/zarch/strmm8x4V.S

@@ -186,7 +186,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,8,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L8x4_BK_Store
@@ -239,7 +239,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,4
       nill LOCAL_VAR1,3 
 #else
-    la LOCAL_VAR1,3(0,0)
+    lghi LOCAL_VAR1,3
     NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x4_BK_Store
@@ -290,7 +290,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x4_BK_Store
@@ -341,7 +341,7 @@ ALIGN_4
        RefreshTempBk LOCAL_VAR1,BK,OFF,1,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/
 #endif 
 jz .L1x4_BK_Store
@@ -423,7 +423,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,8,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L8x2_BK_Store
@@ -475,7 +475,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x2_BK_Store
@@ -525,7 +525,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x2_BK_Store
@@ -575,7 +575,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x2_BK_Store
@@ -655,7 +655,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,8,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L8x1_BK_Store
@@ -708,7 +708,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x1_BK_Store
@@ -757,7 +757,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x1_BK_Store
@@ -807,7 +807,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x1_BK_Store

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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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 struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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

@@ -106,7 +106,7 @@ static FLOAT zasum_kernel_16(BLASLONG n, FLOAT *x) {
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[asum],0"
     : [asum] "=Q"(asum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -36,9 +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"(*(struct { FLOAT x[n * 2]; } *) y),[x] "+&a"(x),[y] "+&a"(y),
+    : "=m"(*(FLOAT (*)[n * 2]) y),[x] "+&a"(x),[y] "+&a"(y),
        [n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x)
+    : "m"(*(const FLOAT (*)[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"(*(struct { FLOAT x[4]; } *) d),[n] "+&r"(n)
-    : [d] "a"(d), "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) y),[y] "a"(y)
+    : "=m"(*(FLOAT (*)[4]) d),[n] "+&r"(n)
+    : [d] "a"(d), "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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

@@ -112,12 +112,12 @@ 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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[8]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n * 2]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[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");
@@ -172,10 +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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-         "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-         "m"(*(const struct { FLOAT x[4]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+         "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+         "m"(*(const FLOAT (*)[4]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27");
 }
@@ -210,9 +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"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-       "m"(*(const struct { FLOAT x[2]; } *) x),[x] "a"(x)
+    : "+m"(*(FLOAT (*)[n * 2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+       "m"(*(const FLOAT (*)[2]) x),[x] "a"(x)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21");
 }
 
@@ -261,8 +261,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) dest),[n] "+&r"(n)
-    : [dest] "a"(dest), "m"(*(const struct { FLOAT x[n * 2]; } *) src),
+    : "+m"(*(FLOAT (*)[n * 2]) dest),[n] "+&r"(n)
+    : [dest] "a"(dest), "m"(*(const FLOAT (*)[n * 2]) src),
        [src] "a"(src),[alpha_r] "Q"(alpha_r),[alpha_i] "Q"(alpha_i)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",

kernel/zarch/zgemv_t_4.c

@@ -141,13 +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"(*(struct { FLOAT x[8]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap2),[ap2] "a"(ap2),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap3),[ap3] "a"(ap3),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[8]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) ap2),[ap2] "a"(ap2),
+       "m"(*(const FLOAT (*)[n * 2]) ap3),[ap3] "a"(ap3),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[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");
@@ -229,11 +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"(*(struct { FLOAT x[4]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap0),[ap0] "a"(ap0),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) ap1),[ap1] "a"(ap1),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[4]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap0),[ap0] "a"(ap0),
+       "m"(*(const FLOAT (*)[n * 2]) ap1),[ap1] "a"(ap1),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23");
 }
@@ -294,10 +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"(*(struct { FLOAT x[2]; } *) y),[n] "+&r"(n)
-    : [y] "a"(y), "m"(*(const struct { FLOAT x[n * 2]; } *) ap),[ap] "a"(ap),
-       "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x),
-       "m"(*(const struct { FLOAT x[2]; } *) alpha),[alpha] "a"(alpha)
+    : "+m"(*(FLOAT (*)[2]) y),[n] "+&r"(n)
+    : [y] "a"(y), "m"(*(const FLOAT (*)[n * 2]) ap),[ap] "a"(ap),
+       "m"(*(const FLOAT (*)[n * 2]) x),[x] "a"(x),
+       "m"(*(const FLOAT (*)[2]) alpha),[alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19");
 }
 

kernel/zarch/zrot.c

@@ -169,8 +169,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),
-       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n * 2]) x),
+       "+m"(*(FLOAT (*)[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",

kernel/zarch/zscal.c

@@ -78,8 +78,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v1", "v16", "v17", "v18", "v19", "v20", "v21",
        "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30",
@@ -128,8 +128,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
@@ -167,8 +167,8 @@ 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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
-    : [x] "a"(x), "m"(*(const struct { FLOAT x[2]; } *) alpha),
+    : "+m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
+    : [x] "a"(x), "m"(*(const FLOAT (*)[2]) alpha),
        [alpha] "a"(alpha)
     : "cc", "r1", "v0", "v16", "v17", "v18", "v19", "v20", "v21", "v22",
        "v23");
@@ -190,7 +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"(*(struct { FLOAT x[n * 2]; } *) x),[n] "+&r"(n)
+    : "=m"(*(FLOAT (*)[n * 2]) x),[n] "+&r"(n)
     : [x] "a"(x)
     : "cc", "r1", "v0");
 }

kernel/zarch/zsum.c

@@ -89,7 +89,7 @@ static FLOAT zsum_kernel_16(BLASLONG n, FLOAT *x) {
     "vfadb   %%v24,%%v24,%%v25\n\t"
     "vsteg   %%v24,%[sum],0"
     : [sum] "=Q"(sum),[n] "+&r"(n)
-    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "m"(*(const FLOAT (*)[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/zswap.c

@@ -99,8 +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"(*(struct { FLOAT x[n * 2]; } *) x),
-       "+m"(*(struct { FLOAT x[n * 2]; } *) y),[n] "+&r"(n)
+    : "+m"(*(FLOAT (*)[n * 2]) x),
+       "+m"(*(FLOAT (*)[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/ztrmm4x4V.S

@@ -196,7 +196,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,4
       nill LOCAL_VAR1,3 
 #else
-    la LOCAL_VAR1,3(0,0)
+    lghi LOCAL_VAR1,3
     NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x4_BK_Store
@@ -256,7 +256,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x4_BK_Store
@@ -307,7 +307,7 @@ ALIGN_4
        RefreshTempBk LOCAL_VAR1,BK,OFF,1,4
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/
 #endif 
 jz .L1x4_BK_Store
@@ -390,7 +390,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x2_BK_Store
@@ -447,7 +447,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x2_BK_Store
@@ -497,7 +497,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,2
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x2_BK_Store
@@ -573,7 +573,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,4,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L4x1_BK_Store
@@ -625,7 +625,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,2,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L2x1_BK_Store
@@ -675,7 +675,7 @@ ALIGN_4
       RefreshTempBk LOCAL_VAR1,BK,OFF,1,1
       nill LOCAL_VAR1,3 
 #else
-la LOCAL_VAR1,3(0,0)
+lghi LOCAL_VAR1,3
 NGR LOCAL_VAR1,BK /*refresh BK*/ 
 #endif
 jz .L1x1_BK_Store

relapack/src/cgbtrf.c

@@ -36,6 +36,7 @@ void RELAPACK_cgbtrf(
         return;
     }
 
+    if (*m == 0 || *n == 0) return;
     // Constant
     const float ZERO[] = { 0., 0. };
 
@@ -56,10 +57,10 @@ void RELAPACK_cgbtrf(
 
     // Allocate work space
     const blasint n1 = CREC_SPLIT(*n);
-    const blasint mWorkl = (kv > n1) ? MAX(1, *m - *kl) : kv;
-    const blasint nWorkl = (kv > n1) ? n1 : kv;
-    const blasint mWorku = (*kl > n1) ? n1 : *kl;
-    const blasint nWorku = (*kl > n1) ? MAX(0, *n - *kl) : *kl;
+    const blasint mWorkl = abs ( (kv > n1) ? MAX(1, *m - *kl) : kv);
+    const blasint nWorkl = abs ( (kv > n1) ? n1 : kv);
+    const blasint mWorku = abs ((*kl > n1) ? n1 : *kl);
+    const blasint nWorku = abs ((*kl > n1) ? MAX(0, *n - *kl) : *kl);
     float *Workl = malloc(mWorkl * nWorkl * 2 * sizeof(float));
     float *Worku = malloc(mWorku * nWorku * 2 * sizeof(float));
     LAPACK(claset)("L", &mWorkl, &nWorkl, ZERO, ZERO, Workl, &mWorkl);
@@ -82,7 +83,7 @@ static void RELAPACK_cgbtrf_rec(
     blasint *info
 ) {
 
-    if (*n <= MAX(CROSSOVER_CGBTRF, 1)) {
+    if (*n <= MAX(CROSSOVER_CGBTRF, 1)|| *n > *kl || *ldAb == 1) {
         // Unblocked
         LAPACK(cgbtf2)(m, n, kl, ku, Ab, ldAb, ipiv, info);
         return;

relapack/src/cgetrf.c

@@ -30,6 +30,8 @@ void RELAPACK_cgetrf(
         return;
     }
 
+    if (*m == 0 || *n == 0) return;
+
     const blasint sn = MIN(*m, *n);
 
     RELAPACK_cgetrf_rec(m, &sn, A, ldA, ipiv, info);
@@ -62,9 +64,11 @@ static void RELAPACK_cgetrf_rec(
     blasint *info
 ) {
 
-    if (*n <= MAX(CROSSOVER_CGETRF, 1)) {
+    if (*m == 0 || *n == 0) return;
+
+    if ( *n <= MAX(CROSSOVER_CGETRF, 1)) {
         // Unblocked
-        LAPACK(cgetf2)(m, n, A, ldA, ipiv, info);
+        LAPACK(cgetrf2)(m, n, A, ldA, ipiv, info);
         return;
     }
 
@@ -96,6 +100,7 @@ static void RELAPACK_cgetrf_rec(
 
     // recursion(A_L, ipiv_T)
     RELAPACK_cgetrf_rec(m, &n1, A_L, ldA, ipiv_T, info);
+    if (*info) return;
     // apply pivots to A_R
     LAPACK(claswp)(&n2, A_R, ldA, iONE, &n1, ipiv_T, iONE);
 

relapack/src/chegst.c

@@ -40,6 +40,8 @@ void RELAPACK_chegst(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower ? 'L' : 'U';
 

relapack/src/chetrf_rook.c

@@ -36,7 +36,7 @@ void RELAPACK_chetrf_rook(
         *info = -2;
     else if (*ldA < MAX(1, *n))
         *info = -4;
-    else if (*lWork < minlWork && *lWork != -1)
+    else if ((*lWork < 1 || *lWork < minlWork) && *lWork != -1)
         *info = -7;
     else if (*lWork == -1) {
         // Work size query
@@ -56,7 +56,7 @@ void RELAPACK_chetrf_rook(
 
     if (*info) {
         const blasint minfo = -*info;
-        LAPACK(xerbla)("CHETRF", &minfo, strlen("CHETRF"));
+        LAPACK(xerbla)("CHETRF_ROOK", &minfo, strlen("CHETRF_ROOK"));
         return;
     }
 

relapack/src/clauum.c

@@ -32,6 +32,8 @@ void RELAPACK_clauum(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower ? 'L' : 'U';
 

relapack/src/cpbtrf.c

@@ -35,6 +35,8 @@ void RELAPACK_cpbtrf(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower ? 'L' : 'U';
 
@@ -43,8 +45,8 @@ void RELAPACK_cpbtrf(
 
     // Allocate work space
     const blasint n1 = CREC_SPLIT(*n);
-    const blasint mWork = (*kd > n1) ? (lower ? *n - *kd : n1) : *kd;
-    const blasint nWork = (*kd > n1) ? (lower ? n1 : *n - *kd) : *kd;
+    const blasint mWork = abs((*kd > n1) ? (lower ? *n - *kd : n1) : *kd);
+    const blasint nWork = abs((*kd > n1) ? (lower ? n1 : *n - *kd) : *kd);
     float *Work = malloc(mWork * nWork * 2 * sizeof(float));
     LAPACK(claset)(uplo, &mWork, &nWork, ZERO, ZERO, Work, &mWork);
 
@@ -64,7 +66,7 @@ static void RELAPACK_cpbtrf_rec(
     blasint *info
 ){
 
-    if (*n <= MAX(CROSSOVER_CPBTRF, 1)) {
+    if (*n <= MAX(CROSSOVER_CPBTRF, 1) || *ldAb==1) {
         // Unblocked
         LAPACK(cpbtf2)(uplo, n, kd, Ab, ldAb, info);
         return;
@@ -148,7 +150,7 @@ static void RELAPACK_cpbtrf_rec(
     }
 
     // recursion(A_BR)
-    if (*kd > n1)
+    if (*kd > n1 && ldA != 0)
         RELAPACK_cpotrf(uplo, &n2, A_BR, ldA, info);
     else
         RELAPACK_cpbtrf_rec(uplo, &n2, kd, Ab_BR, ldAb, Work, ldWork, info);

relapack/src/cpotrf.c

@@ -32,6 +32,8 @@ void RELAPACK_cpotrf(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower ? 'L' : 'U';
 
@@ -46,6 +48,7 @@ static void RELAPACK_cpotrf_rec(
     float *A, const blasint *ldA,
     blasint *info
 ){
+    if (*n == 0) return;
 
     if (*n <= MAX(CROSSOVER_CPOTRF, 1)) {
         // Unblocked

relapack/src/csytrf.c

@@ -36,7 +36,7 @@ void RELAPACK_csytrf(
         *info = -2;
     else if (*ldA < MAX(1, *n))
         *info = -4;
-    else if (*lWork < minlWork && *lWork != -1)
+    else if ((*lWork < 1 || *lWork < minlWork) && *lWork != -1)
         *info = -7;
     else if (*lWork == -1) {
         // Work size query
@@ -67,6 +67,7 @@ void RELAPACK_csytrf(
     blasint nout;
 
     // Recursive kernel
+if (*n != 0)
     RELAPACK_csytrf_rec(&cleanuplo, n, n, &nout, A, ldA, ipiv, cleanWork, n, info);
 
 #if XSYTRF_ALLOW_MALLOC

relapack/src/csytrf_rook.c

@@ -36,7 +36,7 @@ void RELAPACK_csytrf_rook(
         *info = -2;
     else if (*ldA < MAX(1, *n))
         *info = -4;
-    else if (*lWork < minlWork && *lWork != -1)
+    else if ((*lWork < 1 || *lWork < minlWork) && *lWork != -1)
         *info = -7;
     else if (*lWork == -1) {
         // Work size query
@@ -56,7 +56,7 @@ void RELAPACK_csytrf_rook(
 
     if (*info) {
         const blasint minfo = -*info;
-        LAPACK(xerbla)("CSYTRF", &minfo, strlen("CSYTRF"));
+        LAPACK(xerbla)("CSYTRF_ROOK", &minfo, strlen("CSYTRF_ROOK"));
         return;
     }
 

relapack/src/ctgsyl.c

@@ -68,6 +68,13 @@ void RELAPACK_ctgsyl(
         return;
     }
 
+    if ( *m == 0 || *n == 0) {
+      *scale = 1.;
+      if (notran && (*ijob != 0))
+        *dif = 0.;
+      return;
+    }
+
     // Clean char * arguments
     const char cleantrans = notran ? 'N' : 'C';
 

relapack/src/ctrsyl.c

@@ -47,6 +47,11 @@ void RELAPACK_ctrsyl(
         return;
     }
 
+    if (*m == 0 || *n == 0) {
+      *scale = 1.;
+      return;
+    }
+
     // Clean char * arguments
     const char cleantranA = notransA ? 'N' : 'C';
     const char cleantranB = notransB ? 'N' : 'C';

relapack/src/ctrtri.c

@@ -36,6 +36,8 @@ void RELAPACK_ctrtri(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower  ? 'L' : 'U';
     const char cleandiag = nounit ? 'N' : 'U';

relapack/src/dgbtrf.c

@@ -36,6 +36,8 @@ void RELAPACK_dgbtrf(
         return;
     }
 
+    if (*m == 0 || *n == 0) return;
+
     // Constant
     const double ZERO[] = { 0. };
 
@@ -83,7 +85,7 @@ static void RELAPACK_dgbtrf_rec(
     blasint *info
 ) {
 
-    if (*n <= MAX(CROSSOVER_DGBTRF, 1)) {
+    if (*n <= MAX(CROSSOVER_DGBTRF, 1) || *n > *kl || *ldAb == 1) {
         // Unblocked
         LAPACK(dgbtf2)(m, n, kl, ku, Ab, ldAb, ipiv, info);
         return;
@@ -195,6 +197,7 @@ static void RELAPACK_dgbtrf_rec(
     // Worku = A_TRr
     LAPACK(dlacpy)("L", &m1, &n22, A_TRr, ldA, Worku, ldWorku);
     // Worku = A_TL \ Worku
+    if (ldWorku <= 0) return;
     BLAS(dtrsm)("L", "L", "N", "U", &m1, &n22, ONE, A_TL, ldA, Worku, ldWorku);
     // A_TRr = Worku
     LAPACK(dlacpy)("L", &m1, &n22, Worku, ldWorku, A_TRr, ldA);

relapack/src/dgetrf.c

@@ -29,15 +29,16 @@ void RELAPACK_dgetrf(
         return;
     }
 
-    const blasint sn = MIN(*m, *n);
+    if (*m == 0 || *n == 0) return;
 
+    const blasint sn = MIN(*m, *n);
     RELAPACK_dgetrf_rec(m, &sn, A, ldA, ipiv, info);
 
     // Right remainder
     if (*m < *n) {
         // Constants
         const double ONE[] = { 1. };
-        const blasint   iONE[] = { 1. };
+        const blasint   iONE[] = { 1 };
 
         // Splitting
         const blasint rn = *n - *m;
@@ -60,13 +61,11 @@ static void RELAPACK_dgetrf_rec(
     double *A, const blasint *ldA, blasint *ipiv,
     blasint *info
 ) {
-
-    if (*n <= MAX(CROSSOVER_DGETRF, 1)) {
+    if ( *n <= MAX(CROSSOVER_DGETRF, 1)) {
         // Unblocked
-        LAPACK(dgetf2)(m, n, A, ldA, ipiv, info);
+        LAPACK(dgetrf2)(m, n, A, ldA, ipiv, info);
         return;
     }
-
     // Constants
     const double ONE[]  = { 1. };
     const double MONE[] = { -1. };
@@ -95,6 +94,7 @@ static void RELAPACK_dgetrf_rec(
 
     // recursion(A_L, ipiv_T)
     RELAPACK_dgetrf_rec(m, &n1, A_L, ldA, ipiv_T, info);
+    if (*info) return;
     // apply pivots to A_R
     LAPACK(dlaswp)(&n2, A_R, ldA, iONE, &n1, ipiv_T, iONE);
 

relapack/src/dpbtrf.c

@@ -35,6 +35,8 @@ void RELAPACK_dpbtrf(
         return;
     }
 
+    if (*n == 0) return;
+
     // Clean char * arguments
     const char cleanuplo = lower ? 'L' : 'U';
 
@@ -43,8 +45,8 @@ void RELAPACK_dpbtrf(
 
     // Allocate work space
     const blasint n1 = DREC_SPLIT(*n);
-    const blasint mWork = (*kd > n1) ? (lower ? *n - *kd : n1) : *kd;
-    const blasint nWork = (*kd > n1) ? (lower ? n1 : *n - *kd) : *kd;
+    const blasint mWork = abs((*kd > n1) ? (lower ? *n - *kd : n1) : *kd);
+    const blasint nWork = abs((*kd > n1) ? (lower ? n1 : *n - *kd) : *kd);
     double *Work = malloc(mWork * nWork * sizeof(double));
     LAPACK(dlaset)(uplo, &mWork, &nWork, ZERO, ZERO, Work, &mWork);
 
@@ -64,7 +66,7 @@ static void RELAPACK_dpbtrf_rec(
     blasint *info
 ){
 
-    if (*n <= MAX(CROSSOVER_DPBTRF, 1)) {
+    if (*n <= MAX(CROSSOVER_DPBTRF, 1) || *ldAb == 1) {
         // Unblocked
         LAPACK(dpbtf2)(uplo, n, kd, Ab, ldAb, info);
         return;
@@ -148,7 +150,7 @@ static void RELAPACK_dpbtrf_rec(
     }
 
     // recursion(A_BR)
-    if (*kd > n1)
+    if (*kd > n1 && ldA != 0)
         RELAPACK_dpotrf(uplo, &n2, A_BR, ldA, info);
     else
         RELAPACK_dpbtrf_rec(uplo, &n2, kd, Ab_BR, ldAb, Work, ldWork, info);

relapack/src/dsytrf.c

@@ -36,7 +36,7 @@ void RELAPACK_dsytrf(
         *info = -2;
     else if (*ldA < MAX(1, *n))
         *info = -4;
-    else if (*lWork < minlWork && *lWork != -1)
+    else if ((*lWork < 1 || *lWork < minlWork) && *lWork != -1)
         *info = -7;
     else if (*lWork == -1) {
         // Work size query
@@ -67,6 +67,7 @@ void RELAPACK_dsytrf(
     blasint nout;
 
     // Recursive kernel
+if (*n != 0)
     RELAPACK_dsytrf_rec(&cleanuplo, n, n, &nout, A, ldA, ipiv, cleanWork, n, info);
 
 #if XSYTRF_ALLOW_MALLOC

relapack/src/dsytrf_rook.c

@@ -36,7 +36,7 @@ void RELAPACK_dsytrf_rook(
         *info = -2;
     else if (*ldA < MAX(1, *n))
         *info = -4;
-    else if (*lWork < minlWork && *lWork != -1)
+    else if ((*lWork <1 || *lWork < minlWork) && *lWork != -1)
         *info = -7;
     else if (*lWork == -1) {
         // Work size query
@@ -56,7 +56,7 @@ void RELAPACK_dsytrf_rook(
 
     if (*info) {
         const blasint minfo = -*info;
-        LAPACK(xerbla)("DSYTRF", &minfo, strlen("DSYTRF"));
+        LAPACK(xerbla)("DSYTRF_ROOK", &minfo, strlen("DSYTRF_ROOK"));
         return;
     }
 

relapack/src/dtrsyl.c

@@ -49,6 +49,11 @@ void RELAPACK_dtrsyl(
         return;
     }
 
+    if (*m == 0 || *n == 0) {
+        *scale = 1.;
+        return;
+    }
+
     // Clean char * arguments
     const char cleantranA = notransA ? 'N' : (transA ? 'T' : 'C');
     const char cleantranB = notransB ? 'N' : (transB ? 'T' : 'C');

relapack/src/lapack.h

@@ -4,6 +4,13 @@
 extern blasint LAPACK(lsame)(const char *, const char *);
 extern blasint LAPACK(xerbla)(const char *, const blasint *, int);
 
+extern const blasint LAPACK(ilaenv)(const blasint *, const char*, const char*, const blasint* ,  int ,  int,  int );
+
+extern void LAPACK(sgetrf2)(const blasint *, const blasint *, float *, const blasint *, blasint *, blasint *);
+extern void LAPACK(dgetrf2)(const blasint *, const blasint *, double *, const blasint *, blasint *, blasint *);
+extern void LAPACK(cgetrf2)(const blasint *, const blasint *, float *, const blasint *, blasint *, blasint *);
+extern void LAPACK(zgetrf2)(const blasint *, const blasint *, double *, const blasint *, blasint *, blasint *);
+
 extern void LAPACK(slaswp)(const blasint *, float *, const blasint *, const blasint *, const blasint *, const blasint *, const blasint *);
 extern void LAPACK(dlaswp)(const blasint *, double *, const blasint *, const blasint *, const blasint *, const blasint *, const blasint *);
 extern void LAPACK(claswp)(const blasint *, float *, const blasint *, const blasint *, const blasint *, const blasint *, const blasint *);

relapack/src/sgbtrf.c

@@ -35,6 +35,13 @@ void RELAPACK_sgbtrf(
         return;
     }
 
+    if (*m == 0 || *n == 0) return;
+
+    if (*ldAb == 1) {
+        LAPACK(sgbtf2)(m, n, kl, ku, Ab, ldAb, ipiv, info);
+	return;
+    }
+
     // Constant
     const float ZERO[] = { 0. };
 
@@ -82,8 +89,9 @@ static void RELAPACK_sgbtrf_rec(
     blasint *info
 ) {
 
+    if (*m == 0 || *n == 0) return;
 
-    if (*n <= MAX(CROSSOVER_SGBTRF, 1)) {
+    if ( *n <= MAX(CROSSOVER_SGBTRF, 1) || *n > *kl || *ldAb == 1) {
         // Unblocked
         LAPACK(sgbtf2)(m, n, kl, ku, Ab, ldAb, ipiv, info);
         return;
@@ -160,7 +168,7 @@ static void RELAPACK_sgbtrf_rec(
 
     // recursion(Ab_L, ipiv_T)
     RELAPACK_sgbtrf_rec(m, &n1, kl, ku, Ab_L, ldAb, ipiv_T, Workl, ldWorkl, Worku, ldWorku, info);
-
+    if (*info) return;
     // Workl = A_BLb
     LAPACK(slacpy)("U", &m22, &n1, A_BLb, ldA, Workl, ldWorkl);
 
@@ -222,8 +230,8 @@ static void RELAPACK_sgbtrf_rec(
 
     // recursion(Ab_BR, ipiv_B)
 //cause of infinite recursion here ?    
-//      RELAPACK_sgbtrf_rec(&m2, &n2, kl, ku, Ab_BR, ldAb, ipiv_B, Workl, ldWorkl, Worku, ldWorku, info);
-        LAPACK(sgbtf2)(&m2, &n2, kl, ku, Ab_BR, ldAb, ipiv_B, info);
+      RELAPACK_sgbtrf_rec(&m2, &n2, kl, ku, Ab_BR, ldAb, ipiv_B, Workl, ldWorkl, Worku, ldWorku, info);
+//        LAPACK(sgbtf2)(&m2, &n2, kl, ku, Ab_BR, ldAb, ipiv_B, info);
     if (*info)
         *info += n1;
     // shift pivots