OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

Merge pull request #2010 from martin-frbg/issue2009 

Fix declaration of input arguments in x86_64 GEMV, SYMV and DSCAL
tags/v0.3.6^2
Martin Kroeker GitHub 7 years ago
parent
03a2bf2602 ab1630f9fa
commit
63d7bad8a5
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
21 changed files with 73 additions and 73 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +5
    -5
      kernel/x86_64/dgemv_n_4.c
  2. +9
    -9
      kernel/x86_64/dgemv_t_4.c
  3. +1
    -1
      kernel/x86_64/dscal.c
  4. +2
    -2
      kernel/x86_64/dsymv_L_microk_bulldozer-2.c
  5. +2
    -2
      kernel/x86_64/dsymv_L_microk_haswell-2.c
  6. +2
    -2
      kernel/x86_64/dsymv_L_microk_nehalem-2.c
  7. +2
    -2
      kernel/x86_64/dsymv_L_microk_sandy-2.c
  8. +3
    -3
      kernel/x86_64/dsymv_U_microk_bulldozer-2.c
  9. +3
    -3
      kernel/x86_64/dsymv_U_microk_haswell-2.c
  10. +3
    -3
      kernel/x86_64/dsymv_U_microk_nehalem-2.c
  11. +3
    -3
      kernel/x86_64/dsymv_U_microk_sandy-2.c
  12. +7
    -7
      kernel/x86_64/sgemv_n_4.c
  13. +9
    -9
      kernel/x86_64/sgemv_t_4.c
  14. +2
    -2
      kernel/x86_64/ssymv_L_microk_bulldozer-2.c
  15. +2
    -2
      kernel/x86_64/ssymv_L_microk_haswell-2.c
  16. +2
    -2
      kernel/x86_64/ssymv_L_microk_nehalem-2.c
  17. +4
    -4
      kernel/x86_64/ssymv_L_microk_sandy-2.c
  18. +3
    -3
      kernel/x86_64/ssymv_U_microk_bulldozer-2.c
  19. +3
    -3
      kernel/x86_64/ssymv_U_microk_haswell-2.c
  20. +3
    -3
      kernel/x86_64/ssymv_U_microk_nehalem-2.c
  21. +3
    -3
      kernel/x86_64/ssymv_U_microk_sandy-2.c

+ 5
- 5
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" "jnz 1b \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (ap[0]), // 4 "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" "jnz 1b \n\t"


: :
"+r" (i), // 0
"+r" (n) // 1
: :
"r" (i), // 0
"r" (n), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (ap), // 4 "r" (ap), // 4


+ 9
- 9
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" "movsd %%xmm11,8(%2) \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2 "r" (y), // 2
"r" (ap0), // 3 "r" (ap0), // 3
"r" (ap1), // 4 "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" "movsd %%xmm10, (%2) \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2 "r" (y), // 2
"r" (ap), // 3 "r" (ap), // 3
"r" (x) // 4 "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" "jnz 1b \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (&da), // 2 "r" (&da), // 2
"r" (src), // 3 "r" (src), // 3
"r" (dest) // 4 "r" (dest) // 4


+ 1
- 1
kernel/x86_64/dscal.c View File

@@ -136,8 +136,8 @@ static void dscal_kernel_inc_8(BLASLONG n, FLOAT *alpha, FLOAT *x, BLASLONG inc_
"jnz 1b \n\t" "jnz 1b \n\t"


: :
"+r" (n) // 0
: :
"r" (n), // 0
"r" (x), // 1 "r" (x), // 1
"r" (x1), // 2 "r" (x1), // 2
"r" (alpha), // 3 "r" (alpha), // 3


+ 2
- 2
kernel/x86_64/dsymv_L_microk_bulldozer-2.c View File

@@ -113,8 +113,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vmovsd %%xmm3 ,24(%9) \n\t" // save temp2 "vmovsd %%xmm3 ,24(%9) \n\t" // save temp2


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 2
- 2
kernel/x86_64/dsymv_L_microk_haswell-2.c View File

@@ -105,8 +105,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 2
- 2
kernel/x86_64/dsymv_L_microk_nehalem-2.c View File

@@ -108,8 +108,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"movsd %%xmm3 , 24(%9) \n\t" // save temp2 "movsd %%xmm3 , 24(%9) \n\t" // save temp2


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 2
- 2
kernel/x86_64/dsymv_L_microk_sandy-2.c View File

@@ -114,8 +114,8 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 3
- 3
kernel/x86_64/dsymv_U_microk_bulldozer-2.c View File

@@ -106,9 +106,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vmovsd %%xmm3 ,24(%9) \n\t" // save temp2 "vmovsd %%xmm3 ,24(%9) \n\t" // save temp2


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/dsymv_U_microk_haswell-2.c View File

@@ -107,9 +107,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/dsymv_U_microk_nehalem-2.c View File

@@ -101,9 +101,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"movsd %%xmm3 , 24(%9) \n\t" // save temp2 "movsd %%xmm3 , 24(%9) \n\t" // save temp2


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/dsymv_U_microk_sandy-2.c View File

@@ -116,9 +116,9 @@ static void dsymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 7
- 7
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" "jnz 1b \n\t"


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


: :
"+r" (i), // 0
"+r" (n) // 1
: :
"r" (i), // 0
"r" (n), // 1
"r" (src), // 2 "r" (src), // 2
"r" (dest) // 3 "r" (dest) // 3
: "cc", : "cc",


+ 9
- 9
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" "movss %%xmm11,4(%2) \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2 "r" (y), // 2
"r" (ap0), // 3 "r" (ap0), // 3
"r" (ap1), // 4 "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" "movss %%xmm10, (%2) \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (y), // 2 "r" (y), // 2
"r" (ap), // 3 "r" (ap), // 3
"r" (x) // 4 "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" "jnz 1b \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (&da), // 2 "r" (&da), // 2
"r" (src), // 3 "r" (src), // 3
"r" (dest) // 4 "r" (dest) // 4


+ 2
- 2
kernel/x86_64/ssymv_L_microk_bulldozer-2.c View File

@@ -98,8 +98,8 @@ static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vmovss %%xmm3 ,12(%9) \n\t" // save temp2 "vmovss %%xmm3 ,12(%9) \n\t" // save temp2


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 2
- 2
kernel/x86_64/ssymv_L_microk_haswell-2.c View File

@@ -99,8 +99,8 @@ static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 2
- 2
kernel/x86_64/ssymv_L_microk_nehalem-2.c View File

@@ -113,8 +113,8 @@ static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, F
"movss %%xmm3 , 12(%9) \n\t" // save temp2 "movss %%xmm3 , 12(%9) \n\t" // save temp2


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 4
- 4
kernel/x86_64/ssymv_L_microk_sandy-2.c View File

@@ -109,8 +109,8 @@ static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
@@ -217,8 +217,8 @@ static void ssymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (from), // 0
"+r" (from) // 0
:
"r" (to), // 1 "r" (to), // 1
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3


+ 3
- 3
kernel/x86_64/ssymv_U_microk_bulldozer-2.c View File

@@ -90,9 +90,9 @@ static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vmovss %%xmm3 ,12(%9) \n\t" // save temp2 "vmovss %%xmm3 ,12(%9) \n\t" // save temp2


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/ssymv_U_microk_haswell-2.c View File

@@ -112,9 +112,9 @@ static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/ssymv_U_microk_nehalem-2.c View File

@@ -106,9 +106,9 @@ static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"movss %%xmm3 , 12(%9) \n\t" // save temp2 "movss %%xmm3 , 12(%9) \n\t" // save temp2


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


+ 3
- 3
kernel/x86_64/ssymv_U_microk_sandy-2.c View File

@@ -120,9 +120,9 @@ static void ssymv_kernel_4x4(BLASLONG n, FLOAT *a0, FLOAT *a1, FLOAT *a2, FLOAT
"vzeroupper \n\t" "vzeroupper \n\t"


: :
:
"r" (i), // 0
"r" (n), // 1
"+r" (i), // 0
"+r" (n) // 1
:
"r" (x), // 2 "r" (x), // 2
"r" (y), // 3 "r" (y), // 3
"r" (a0), // 4 "r" (a0), // 4


Write Preview
Loading…
Cancel
Save