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

ARM64: Convert all labels to local labels

Browse Source 
While debugging/profiling applications using perf or other tools, the
kernels appear scattered in the profile reports. This is because the labels
within the kernels are not local and each label is shown as a separate
function.

To avoid this, all the labels within the kernels are changed to local
labels.
This commit is contained in:
Ashwin Sekhar T K
2017-10-24 10:47:11 +00:00
parent 627133f9ad
commit a0128aa489
50 changed files with 4469 additions and 4469 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
kernel/arm64/dgemm_ncopy_4.S
View File

@@ -192,14 +192,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
lsl LDA, LDA, #3 // LDA = LDA * SIZE
dgemm_ncopy_L4_BEGIN:
.Ldgemm_ncopy_L4_BEGIN:
asr J, N, #2 // J = N / 4
cmp J, #0
ble dgemm_ncopy_L2_BEGIN
ble .Ldgemm_ncopy_L2_BEGIN
.align 5
dgemm_ncopy_L4_M4_BEGIN:
.Ldgemm_ncopy_L4_M4_BEGIN:
mov A01, A00
add A02, A01, LDA
@@ -209,128 +209,128 @@ dgemm_ncopy_L4_M4_BEGIN:
asr I, M, #2 // I = M / 4
cmp I, #0
ble dgemm_ncopy_L4_M4_40
ble .Ldgemm_ncopy_L4_M4_40
.align 5
dgemm_ncopy_L4_M4_20:
.Ldgemm_ncopy_L4_M4_20:
COPY4x4
subs I , I , #1
bne dgemm_ncopy_L4_M4_20
bne .Ldgemm_ncopy_L4_M4_20
dgemm_ncopy_L4_M4_40:
.Ldgemm_ncopy_L4_M4_40:
and I, M , #3
cmp I, #0
ble dgemm_ncopy_L4_M4_END
ble .Ldgemm_ncopy_L4_M4_END
.align 5
dgemm_ncopy_L4_M4_60:
.Ldgemm_ncopy_L4_M4_60:
COPY1x4
subs I , I , #1
bne dgemm_ncopy_L4_M4_60
bne .Ldgemm_ncopy_L4_M4_60
dgemm_ncopy_L4_M4_END:
.Ldgemm_ncopy_L4_M4_END:
subs J , J, #1 // j--
bne dgemm_ncopy_L4_M4_BEGIN
bne .Ldgemm_ncopy_L4_M4_BEGIN
/*********************************************************************************************/
dgemm_ncopy_L2_BEGIN:
.Ldgemm_ncopy_L2_BEGIN:
tst N, #3
ble dgemm_ncopy_L999
ble .Ldgemm_ncopy_L999
tst N, #2
ble dgemm_ncopy_L1_BEGIN
ble .Ldgemm_ncopy_L1_BEGIN
dgemm_ncopy_L2_M4_BEGIN:
.Ldgemm_ncopy_L2_M4_BEGIN:
mov A01, A00
add A02, A01, LDA
add A00, A02, LDA
asr I, M, #2 // I = M / 4
cmp I, #0
ble dgemm_ncopy_L2_M4_40
ble .Ldgemm_ncopy_L2_M4_40
.align 5
dgemm_ncopy_L2_M4_20:
.Ldgemm_ncopy_L2_M4_20:
COPY4x2
subs I , I , #1
bne dgemm_ncopy_L2_M4_20
bne .Ldgemm_ncopy_L2_M4_20
dgemm_ncopy_L2_M4_40:
.Ldgemm_ncopy_L2_M4_40:
and I, M , #3
cmp I, #0
ble dgemm_ncopy_L2_M4_END
ble .Ldgemm_ncopy_L2_M4_END
.align 5
dgemm_ncopy_L2_M4_60:
.Ldgemm_ncopy_L2_M4_60:
COPY1x2
subs I , I , #1
bne dgemm_ncopy_L2_M4_60
bne .Ldgemm_ncopy_L2_M4_60
dgemm_ncopy_L2_M4_END:
.Ldgemm_ncopy_L2_M4_END:
/*********************************************************************************************/
dgemm_ncopy_L1_BEGIN:
.Ldgemm_ncopy_L1_BEGIN:
tst N, #1
ble dgemm_ncopy_L999
ble .Ldgemm_ncopy_L999
dgemm_ncopy_L1_M4_BEGIN:
.Ldgemm_ncopy_L1_M4_BEGIN:
mov A01, A00
asr I, M, #2 // I = M / 4
cmp I, #0
ble dgemm_ncopy_L1_M4_40
ble .Ldgemm_ncopy_L1_M4_40
.align 5
dgemm_ncopy_L1_M4_20:
.Ldgemm_ncopy_L1_M4_20:
COPY4x1
subs I , I , #1
bne dgemm_ncopy_L1_M4_20
bne .Ldgemm_ncopy_L1_M4_20
dgemm_ncopy_L1_M4_40:
.Ldgemm_ncopy_L1_M4_40:
and I, M , #3
cmp I, #0
ble dgemm_ncopy_L1_M4_END
ble .Ldgemm_ncopy_L1_M4_END
.align 5
dgemm_ncopy_L1_M4_60:
.Ldgemm_ncopy_L1_M4_60:
COPY1x1
subs I , I , #1
bne dgemm_ncopy_L1_M4_60
bne .Ldgemm_ncopy_L1_M4_60
dgemm_ncopy_L1_M4_END:
.Ldgemm_ncopy_L1_M4_END:
dgemm_ncopy_L999:
.Ldgemm_ncopy_L999:
mov x0, #0
RESTORE_REGS