Browse Source

small optimizations on dot-kernels

tags/v0.2.9.rc1
wernsaar 12 years ago
parent
commit
f750103336
4 changed files with 22 additions and 12 deletions
  1. +2
    -2
      kernel/arm/cdot_vfpv3.S
  2. +16
    -6
      kernel/arm/ddot_vfpv3.S
  3. +2
    -2
      kernel/arm/sdot_vfpv3.S
  4. +2
    -2
      kernel/arm/zdot_vfpv3.S

+ 2
- 2
kernel/arm/cdot_vfpv3.S View File

@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/

/**************************************************************************************
* 2013/11/08 Saar
* 2013/11/11 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
@@ -56,7 +56,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define Y r6
#define INC_Y r7

#define X_PRE 256
#define X_PRE 512

/**************************************************************************************
* Macro definitions


+ 16
- 6
kernel/arm/ddot_vfpv3.S View File

@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/

/**************************************************************************************
* 2013/11/07 Saar
* 2013/11/11 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
@@ -56,7 +56,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define Y r6
#define INC_Y r7

#define X_PRE 256
#define X_PRE 512

/**************************************************************************************
* Macro definitions
@@ -65,14 +65,18 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.macro KERNEL_F4

pld [ X, #X_PRE ]
fldmiad X!, { d8 }
pld [ Y, #X_PRE ]
fldmiad X!, { d8 - d9 }
fldmiad Y!, { d4 - d5}
fldmiad Y!, { d4 }
fldmiad Y!, { d5 }
fmacd d0 , d4, d8
fldmiad X!, { d10 - d11 }
fldmiad X!, { d9 }
fldmiad Y!, { d6 }
fmacd d1 , d5, d9
fldmiad Y!, { d6 - d7 }
fldmiad X!, { d10 }
fldmiad X!, { d11 }
fmacd d0 , d6, d10
fldmiad Y!, { d7 }
fmacd d1 , d7, d11

.endm
@@ -173,6 +177,12 @@ ddot_kernel_F_BEGIN:

ddot_kernel_F4:

KERNEL_F4

subs I, I, #1
ble ddot_kernel_F1


KERNEL_F4

subs I, I, #1


+ 2
- 2
kernel/arm/sdot_vfpv3.S View File

@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/

/**************************************************************************************
* 2013/11/07 Saar
* 2013/11/11 Saar
* BLASTEST : OK
* CTEST : OK (no test for dsdot)
* TEST : OK (no test for dsdot)
@@ -56,7 +56,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define Y r6
#define INC_Y r7

#define X_PRE 256
#define X_PRE 512

/**************************************************************************************
* Macro definitions


+ 2
- 2
kernel/arm/zdot_vfpv3.S View File

@@ -26,7 +26,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/

/**************************************************************************************
* 2013/11/08 Saar
* 2013/11/11 Saar
* BLASTEST : OK
* CTEST : OK
* TEST : OK
@@ -56,7 +56,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define Y r6
#define INC_Y r7

#define X_PRE 256
#define X_PRE 512

/**************************************************************************************
* Macro definitions


Loading…
Cancel
Save