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

Ref #380: lowered stack usage for piledriver and bulldozer kernels

This commit is contained in:
wernsaar 2014-06-19 14:02:14 +02:00
parent a19d209005
commit 73545a79cd
8 changed files with 12 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
kernel/x86_64/cgemm_kernel_4x2_bulldozer.S
View File

@ -79,8 +79,7 @@
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)

4
kernel/x86_64/cgemm_kernel_4x2_piledriver.S
View File

@ -104,8 +104,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 256*8*4
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)
@ -116,7 +115,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define KK 72(%rsp)
#define KKK 80(%rsp)
#define BUFFER1 128(%rsp)
#define BUFFER2 LB2_OFFSET+128(%rsp)
#if defined(OS_WINDOWS)
#if L_BUFFER_SIZE > 16384

4
kernel/x86_64/dgemm_kernel_8x2_bulldozer.S
View File

@ -148,8 +148,8 @@
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define LB2_OFFSET 4096
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)

4
kernel/x86_64/dgemm_kernel_8x2_piledriver.S
View File

@ -105,8 +105,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define LB2_OFFSET 4096
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)

4
kernel/x86_64/sgemm_kernel_16x2_bulldozer.S
View File

@ -78,8 +78,8 @@
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define LB2_OFFSET 4096
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)

4
kernel/x86_64/sgemm_kernel_16x2_piledriver.S
View File

@ -105,8 +105,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define LB2_OFFSET 4096
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)

4
kernel/x86_64/zgemm_kernel_2x2_bulldozer.S
View File

@ -79,8 +79,7 @@
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 8192
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)
@ -91,7 +90,6 @@
#define KK 72(%rsp)
#define KKK 80(%rsp)
#define BUFFER1 128(%rsp)
#define BUFFER2 LB2_OFFSET+128(%rsp)
#if defined(OS_WINDOWS)
#if L_BUFFER_SIZE > 16384

4
kernel/x86_64/zgemm_kernel_2x2_piledriver.S
View File

@ -104,8 +104,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#define L_BUFFER_SIZE 512*8*4
#define LB2_OFFSET 512*8*2
#define L_BUFFER_SIZE 256*8*4
#define Ndiv6 24(%rsp)
#define Nmod6 32(%rsp)
@ -116,7 +115,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define KK 72(%rsp)
#define KKK 80(%rsp)
#define BUFFER1 128(%rsp)
#define BUFFER2 LB2_OFFSET+128(%rsp)
#if defined(OS_WINDOWS)
#if L_BUFFER_SIZE > 16384