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

Fix declaration of input arguments in the x86_64 s/dGEMV_T and s/dGEMV_N kernels 

Arguments 0 and 1 need to be tagged as both input and output
This commit is contained in:
Martin Kroeker 2019-02-12 15:33:48 +01:00 committed by GitHub
parent 69edc5bbe7
commit dc6ac9eab0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 30 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
kernel/x86_64/dgemv_n_4.c
View File

@ -111,9 +111,9 @@ static void dgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
"jnz 1b \n\t"
:
"+r" (i), // 0
"+r" (n) // 1
:
"r" (i), // 0
"r" (n), // 1
"r" (x), // 2
"r" (y), // 3
"r" (ap[0]), // 4
@ -166,9 +166,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
"jnz 1b \n\t"
:
"+r" (i), // 0
"+r" (n) // 1
:
"r" (i), // 0
"r" (n), // 1
"r" (x), // 2
"r" (y), // 3
"r" (ap), // 4

18
kernel/x86_64/dgemv_t_4.c
View File

@ -127,9 +127,9 @@ static void dgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
"movsd %%xmm11,8(%2) \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2
"r" (ap0), // 3
"r" (ap1), // 4
@ -195,9 +195,9 @@ static void dgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
"movsd %%xmm10, (%2) \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2
"r" (ap), // 3
"r" (x) // 4
@ -259,9 +259,9 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
"jnz 1b \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (&da), // 2
"r" (src), // 3
"r" (dest) // 4

14
kernel/x86_64/sgemv_n_4.c
View File

@ -149,9 +149,9 @@ static void sgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT
"jnz 1b \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2
"r" (y), // 3
"r" (ap[0]), // 4
@ -223,9 +223,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y, FLOAT *a
"3: \n\t"
:
"+r" (i), // 0
"+r" (n1) // 1
:
"r" (i), // 0
"r" (n1), // 1
"r" (x), // 2
"r" (y), // 3
"r" (ap), // 4
@ -277,9 +277,9 @@ static void add_y(BLASLONG n, FLOAT *src, FLOAT *dest, BLASLONG inc_dest)
"jnz 1b \n\t"
:
"+r" (i), // 0
"+r" (n) // 1
:
"r" (i), // 0
"r" (n), // 1
"r" (src), // 2
"r" (dest) // 3
: "cc",

18
kernel/x86_64/sgemv_t_4.c
View File

@ -139,9 +139,9 @@ static void sgemv_kernel_4x2(BLASLONG n, FLOAT *ap0, FLOAT *ap1, FLOAT *x, FLOAT
"movss %%xmm11,4(%2) \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2
"r" (ap0), // 3
"r" (ap1), // 4
@ -208,9 +208,9 @@ static void sgemv_kernel_4x1(BLASLONG n, FLOAT *ap, FLOAT *x, FLOAT *y)
"movss %%xmm10, (%2) \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2
"r" (ap), // 3
"r" (x) // 4
@ -272,9 +272,9 @@ static void add_y(BLASLONG n, FLOAT da , FLOAT *src, FLOAT *dest, BLASLONG inc_d
"jnz 1b \n\t"
:
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (&da), // 2
"r" (src), // 3
"r" (dest) // 4