floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

s390x: avoid variable-length arrays in struct for asm operands

Browse Source 
... since it is not required and clang does not support that gcc
extension. Instead, use a variable-length array directly for these
operands.

Note that, while the actual inline assembly code does not directly use
these memory operands, they serve to inform the compiler that it cannot
reorder reads or writes to/from the input and output data across the
inline asm statements.

Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
This commit is contained in:
Marius Hillenbrand
2020-09-01 12:08:05 +02:00
parent b9b3265ec8
commit 87e5bbd887
71 changed files with 212 additions and 212 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
kernel/zarch/zgemv_t_4.c
View File

@@ -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");
}