OSchip
/
OpenBLAS
Not watched
1
0
Fork 0
Code
Releases 66 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
9504 Commits
51 Branches
78 MB
0 Download
Tree: 114316f361
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '114316f361'
${ noResults }
OpenBLAS/kernel/arm64/sgemm_kernel_4x4.S

sgemm_kernel_4x4.S 30 kB
Raw Normal View History

Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
Cortex-A57: Fix clang compilation errors
9 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Cortex-A57: Fix clang compilation errors
9 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
Improve the sgemm kernel for CORTEXA57
10 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
​
​
​
​
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
ARM64: Convert all labels to local labels 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.
8 years ago
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
ARM64: Convert all labels to local labels 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.
8 years ago
Improve the sgemm kernel for CORTEXA57
10 years ago
​
​
​
​
​
​
​
​
​
​
​
​
# The first commit's message is: Optimizations for APM's xgene-1 (aarch64). 1) general system updates to support armv8 better. Make all did not work, one needed to supply TARGET=ARMV8. 2) sgem 4x4 kernel in assembler using SIMD, and configuration changes to use it. 3) strmm 4x4 kernel in C. Since the sgem kernel does 4x4, the trmm kernel must also do 4xN. Added Dave Nuechterlein to the contributors list.
11 years ago
​
​
​
 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563 
  1. /*******************************************************************************

  2. Copyright (c) 2015, The OpenBLAS Project

  3. All rights reserved.

  4. Redistribution and use in source and binary forms, with or without

  5. modification, are permitted provided that the following conditions are

  6. met:

  7. 1. Redistributions of source code must retain the above copyright

  8. notice, this list of conditions and the following disclaimer.

  9. 2. Redistributions in binary form must reproduce the above copyright

  10. notice, this list of conditions and the following disclaimer in

  11. the documentation and/or other materials provided with the

  12. distribution.

  13. 3. Neither the name of the OpenBLAS project nor the names of

  14. its contributors may be used to endorse or promote products

  15. derived from this software without specific prior written permission.

  16. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"

  17. AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE

  18. IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE

  19. ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE

  20. LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL

  21. DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR

  22. SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER

  23. CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,

  24. OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE

  25. USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

  26. *******************************************************************************/

  27. 

  28. #define ASSEMBLER

  29. #include "common.h"

  30. 

  31. /*                   X0          X1          X2          s0        X3        x4       x5           x6 */

  32. /*int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha0,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc */

  33. 

  34. #define origM		x0

  35. #define origN		x1

  36. #define origK		x2

  37. #define origPA		x3

  38. #define origPB		x4

  39. #define pC		x5

  40. #define LDC		x6

  41. #define temp		x7

  42. #define counterL	x8

  43. #define counterI	x9

  44. #define counterJ	x10

  45. #define pB		x11

  46. #define pCRow0		x12

  47. #define pCRow1		x13

  48. #define pCRow2		x14

  49. #define pA_0		x15

  50. #define pA_1		x16

  51. #define pA_2		x17

  52. #define pA_3		x18

  53. 

  54. 

  55. #define alpha0		s10

  56. #define alphaV0		v10.s[0]

  57. #define alpha1		s11

  58. #define alphaV1		v11.s[0]

  59. #define alpha2		s14

  60. #define alphaV2		v14.s[0]

  61. #define alpha3		s15

  62. #define alphaV3		v15.s[0]

  63. 

  64. // 00 origM

  65. // 01 origN

  66. // 02 origK

  67. // 03 origPA

  68. // 04 origPB

  69. // 05 pC

  70. // 06 origLDC -> LDC

  71. // 07 offset -> temp

  72. // 08 counterL

  73. // 09 counterI

  74. // 10 counterJ

  75. // 11 pB

  76. // 12 pCRow0

  77. // 13 pCRow1

  78. // 14 pCRow2

  79. // 15 pA_0

  80. // 16 pA_1

  81. // 17 pA_2

  82. // 18 must save pA_3

  83. // 19 must save

  84. // 20 must save

  85. // 21 must save

  86. // 22 must save

  87. // 23 must save

  88. // 24 must save

  89. // 25 must save

  90. // 26 must save

  91. // 27 must save

  92. // 28 must save

  93. // 29 frame

  94. // 30 link

  95. // 31 sp

  96. 

  97. /***************************** FOR 16x4 ***************************************/

  98. //v00 ALPHA -> pA00_0, pA01_0, pA02_0, pA03_0

  99. //v01 pA10_0, pA11_0, pA12_0, pA13_0

  100. //v02 pA00_1, pA01_1, pA02_1, pA03_1

  101. //v03 pA10_1, pA11_1, pA12_1, pA13_1

  102. //v04 pA00_2, pA01_2, pA02_2, pA03_2

  103. //v05 pA10_2, pA11_2, pA12_2, pA13_2

  104. //v06 pA00_3, pA01_3, pA02_3, pA03_3

  105. //v07 pA10_3, pA11_3, pA12_3, pA13_3

  106. //v08 must save pB00, pB01, pB02, pB03

  107. //v09 must save

  108. //v10 must save ALPHA0

  109. //v11 must save ALPHA1

  110. //v12 must save pB10, pB11, pB12, pB13

  111. //v13 must save

  112. //v14 must save ALPHA2

  113. //v15 must save ALPHA3

  114. //v16 must save C00_0, C01_0, C02_0, C03_0

  115. //v17 must save C10_0, C11_0, C12_0, C13_0

  116. //v18 C20_0, C21_0, C22_0, C23_0

  117. //v19 C30_0, C31_0, C32_0, C33_0

  118. //v20 C00_1, C01_1, C02_1, C03_1

  119. //v21 C10_1, C11_1, C12_1, C13_1

  120. //v22 C20_1, C21_1, C22_1, C23_1

  121. //v23 C30_1, C31_1, C32_1, C33_1

  122. //v24 C00_2, C01_2, C02_2, C03_2

  123. //v25 C10_2, C11_2, C12_2, C13_2

  124. //v26 C20_2, C21_2, C22_2, C23_2

  125. //v27 C30_2, C31_2, C32_2, C33_2

  126. //v28 C00_3, C01_3, C02_3, C03_3

  127. //v29 C10_3, C11_3, C12_3, C13_3

  128. //v30 C20_3, C21_3, C22_3, C23_3

  129. //v31 C30_3, C31_3, C32_3, C33_3

  130. 

  131. /***************************** EXCEPT FOR 16x4 ********************************/

  132. //v00 ALPHA -> pA00, pA01

  133. //v01 pA02, pA03

  134. //v02 ppA00, ppA01

  135. //v03 ppA02, ppA03

  136. //v04 pA10, pA11

  137. //v05 pA12, pA13

  138. //v06 ppA10, ppA11

  139. //v07 ppA12, ppA13

  140. //v08 must save pB00, pB01

  141. //v09 must save pB02, pB03

  142. //v10 must save ALPHA0

  143. //v11 must save ALPHA1

  144. //v12 must save pB10, pB11

  145. //v13 must save pB12, pB13

  146. //v14 must save ALPHA2

  147. //v15 must save ALPHA3

  148. //v16 must save C00, C01

  149. //v17 must save C02, C03

  150. //v18 ppC00, ppC01

  151. //v19 ppC02, ppC03

  152. //v20 C10, C11

  153. //v21 C12, C13

  154. //v22 ppC10, ppC11

  155. //v23 ppC12, ppC13

  156. //v24 C20, C21

  157. //v25 C22, C23

  158. //v26 ppC20, ppC21

  159. //v27 ppC22, ppC23

  160. //v28 C30, C31

  161. //v29 C32, C33

  162. //v30 ppC30, ppC31

  163. //v31 ppC32, ppC33

  164. 

  165. /*******************************************************************************

  166. * Macro definitions

  167. *******************************************************************************/

  168. 

  169. .macro INIT16x4

  170. 	fmov		s16, wzr

  171. 	fmov		s17, s16

  172. 	fmov		s18, s17

  173. 	fmov		s19, s16

  174. 	fmov		s20, s17

  175. 	fmov		s21, s16

  176. 	fmov		s22, s17

  177. 	fmov		s23, s16

  178. 	fmov		s24, s17

  179. 	fmov		s25, s16

  180. 	fmov		s26, s17

  181. 	fmov		s27, s16

  182. 	fmov		s28, s17

  183. 	fmov		s29, s16

  184. 	fmov		s30, s17

  185. 	fmov		s31, s16

  186. .endm

  187. 

  188. .macro KERNEL16x4_I

  189. 	ld1	{v8.4s}, [pB]

  190. 	add	pB, pB, #16

  191. 

  192. 	ld1	{v0.4s}, [pA_0]

  193. 	add	pA_0, pA_0, #16

  194. 

  195. 	fmul	v16.4s, v0.4s, v8.s[0]

  196. 	fmul	v20.4s, v0.4s, v8.s[1]

  197. 

  198. 	ld1	{v2.4s}, [pA_1]

  199. 	add	pA_1, pA_1, #16

  200. 

  201. 	fmul	v24.4s, v0.4s, v8.s[2]

  202. 	fmul	v28.4s, v0.4s, v8.s[3]

  203. 

  204. 	ld1	{v4.4s}, [pA_2]

  205. 	add	pA_2, pA_2, #16

  206. 

  207. 	fmul	v17.4s, v2.4s, v8.s[0]

  208. 	fmul	v21.4s, v2.4s, v8.s[1]

  209. 

  210. 	ld1	{v6.4s}, [pA_3]

  211. 	add	pA_3, pA_3, #16

  212. 

  213. 	fmul	v25.4s, v2.4s, v8.s[2]

  214. 	fmul	v29.4s, v2.4s, v8.s[3]

  215. 

  216. 	ld1	{v12.4s}, [pB]		// for next round

  217. 	add	pB, pB, #16

  218. 

  219. 	fmul	v18.4s, v4.4s, v8.s[0]

  220. 	fmul	v19.4s, v6.4s, v8.s[0]

  221. 

  222. 	ld1	{v1.4s}, [pA_0]		// for next round

  223. 	add	pA_0, pA_0, #16

  224. 

  225. 	fmul	v22.4s, v4.4s, v8.s[1]

  226. 	fmul	v23.4s, v6.4s, v8.s[1]

  227. 

  228. 	ld1	{v3.4s}, [pA_1]		// for next round

  229. 	add	pA_1, pA_1, #16

  230. 

  231. 	fmul	v26.4s, v4.4s, v8.s[2]

  232. 	fmul	v27.4s, v6.4s, v8.s[2]

  233. 

  234. 	ld1	{v5.4s}, [pA_2]		// for next round

  235. 	add	pA_2, pA_2, #16

  236. 

  237. 	fmul	v30.4s, v4.4s, v8.s[3]

  238. 	fmul	v31.4s, v6.4s, v8.s[3]

  239. 

  240. 	ld1	{v7.4s}, [pA_3]		// for next round

  241. 	add	pA_3, pA_3, #16

  242. .endm

  243. 

  244. .macro KERNEL16x4_M2

  245. 	fmla	v16.4s, v1.4s, v12.s[0]

  246. 	fmla	v17.4s, v3.4s, v12.s[0]

  247. 

  248. 	ld1	{v8.4s}, [pB]		// for next round

  249. 	add	pB, pB, #16

  250. 

  251. 	fmla	v18.4s, v5.4s, v12.s[0]

  252. 	fmla	v19.4s, v7.4s, v12.s[0]

  253. 

  254. 	ld1	{v0.4s}, [pA_0]		// for next round

  255. 	add	pA_0, pA_0, #16

  256. 

  257. 	fmla	v20.4s, v1.4s, v12.s[1]

  258. 	fmla	v21.4s, v3.4s, v12.s[1]

  259. 

  260. 	ld1	{v2.4s}, [pA_1]		// for next round

  261. 	add	pA_1, pA_1, #16

  262. 

  263. 	fmla	v22.4s, v5.4s, v12.s[1]

  264. 	fmla	v23.4s, v7.4s, v12.s[1]

  265. 

  266. 	ld1	{v4.4s}, [pA_2]		// for next round

  267. 	add	pA_2, pA_2, #16

  268. 

  269. 	fmla	v24.4s, v1.4s, v12.s[2]

  270. 	fmla	v25.4s, v3.4s, v12.s[2]

  271. 

  272. 	ld1	{v6.4s}, [pA_3]		// for next round

  273. 	add	pA_3, pA_3, #16

  274. 

  275. 	fmla	v26.4s, v5.4s, v12.s[2]

  276. 	fmla	v27.4s, v7.4s, v12.s[2]

  277. 

  278. 	prfm	PLDL1KEEP, [pA_2, #512]

  279. 

  280. 	fmla	v28.4s, v1.4s, v12.s[3]

  281. 	fmla	v29.4s, v3.4s, v12.s[3]

  282. 

  283. 	prfm	PLDL1KEEP, [pA_3, #512]

  284. 

  285. 	fmla	v30.4s, v5.4s, v12.s[3]

  286. 	fmla	v31.4s, v7.4s, v12.s[3]

  287. 

  288. 	prfm	PLDL1KEEP, [pB, #512]

  289. .endm

  290. 

  291. .macro KERNEL16x4_M1

  292. 	fmla	v16.4s, v0.4s, v8.s[0]

  293. 	fmla	v17.4s, v2.4s, v8.s[0]

  294. 

  295. 	ld1	{v12.4s}, [pB]		// for next round

  296. 	add	pB, pB, #16

  297. 

  298. 	fmla	v18.4s, v4.4s, v8.s[0]

  299. 	fmla	v19.4s, v6.4s, v8.s[0]

  300. 

  301. 	ld1	{v1.4s}, [pA_0]		// for next round

  302. 	add	pA_0, pA_0, #16

  303. 

  304. 	fmla	v20.4s, v0.4s, v8.s[1]

  305. 	fmla	v21.4s, v2.4s, v8.s[1]

  306. 

  307. 	ld1	{v3.4s}, [pA_1]		// for next round

  308. 	add	pA_1, pA_1, #16

  309. 

  310. 	fmla	v22.4s, v4.4s, v8.s[1]

  311. 	fmla	v23.4s, v6.4s, v8.s[1]

  312. 

  313. 	ld1	{v5.4s}, [pA_2]		// for next round

  314. 	add	pA_2, pA_2, #16

  315. 

  316. 	fmla	v24.4s, v0.4s, v8.s[2]

  317. 	fmla	v25.4s, v2.4s, v8.s[2]

  318. 

  319. 	ld1	{v7.4s}, [pA_3]		// for next round

  320. 	add	pA_3, pA_3, #16

  321. 

  322. 	fmla	v26.4s, v4.4s, v8.s[2]

  323. 	fmla	v27.4s, v6.4s, v8.s[2]

  324. 

  325. 	prfm	PLDL1KEEP, [pA_0, #512]

  326. 

  327. 	fmla	v28.4s, v0.4s, v8.s[3]

  328. 	fmla	v29.4s, v2.4s, v8.s[3]

  329. 

  330. 	prfm	PLDL1KEEP, [pA_1, #512]

  331. 

  332. 	fmla	v30.4s, v4.4s, v8.s[3]

  333. 	fmla	v31.4s, v6.4s, v8.s[3]

  334. .endm

  335. 

  336. .macro KERNEL16x4_E

  337. 	fmla	v16.4s, v1.4s, v12.s[0]

  338. 	fmla	v17.4s, v3.4s, v12.s[0]

  339. 	fmla	v18.4s, v5.4s, v12.s[0]

  340. 	fmla	v19.4s, v7.4s, v12.s[0]

  341. 	fmla	v20.4s, v1.4s, v12.s[1]

  342. 	fmla	v21.4s, v3.4s, v12.s[1]

  343. 	fmla	v22.4s, v5.4s, v12.s[1]

  344. 	fmla	v23.4s, v7.4s, v12.s[1]

  345. 	fmla	v24.4s, v1.4s, v12.s[2]

  346. 	fmla	v25.4s, v3.4s, v12.s[2]

  347. 	fmla	v26.4s, v5.4s, v12.s[2]

  348. 	fmla	v27.4s, v7.4s, v12.s[2]

  349. 	fmla	v28.4s, v1.4s, v12.s[3]

  350. 	fmla	v29.4s, v3.4s, v12.s[3]

  351. 	fmla	v30.4s, v5.4s, v12.s[3]

  352. 	fmla	v31.4s, v7.4s, v12.s[3]

  353. .endm

  354. 

  355. .macro KERNEL16x4_SUB

  356. 	ld1	{v8.4s}, [pB]

  357. 	add	pB, pB, #16

  358. 

  359. 	ld1	{v0.4s}, [pA_0]

  360. 	add	pA_0, pA_0, #16

  361. 

  362. 	fmla	v16.4s, v0.4s, v8.s[0]

  363. 	fmla	v20.4s, v0.4s, v8.s[1]

  364. 	fmla	v24.4s, v0.4s, v8.s[2]

  365. 	fmla	v28.4s, v0.4s, v8.s[3]

  366. 

  367. 	ld1	{v2.4s}, [pA_1]

  368. 	add	pA_1, pA_1, #16

  369. 

  370. 	fmla	v17.4s, v2.4s, v8.s[0]

  371. 	fmla	v21.4s, v2.4s, v8.s[1]

  372. 	fmla	v25.4s, v2.4s, v8.s[2]

  373. 	fmla	v29.4s, v2.4s, v8.s[3]

  374. 

  375. 	ld1	{v4.4s}, [pA_2]

  376. 	add	pA_2, pA_2, #16

  377. 

  378. 	fmla	v18.4s, v4.4s, v8.s[0]

  379. 	fmla	v22.4s, v4.4s, v8.s[1]

  380. 	fmla	v26.4s, v4.4s, v8.s[2]

  381. 	fmla	v30.4s, v4.4s, v8.s[3]

  382. 

  383. 	ld1	{v6.4s}, [pA_3]

  384. 	add	pA_3, pA_3, #16

  385. 

  386. 	fmla	v19.4s, v6.4s, v8.s[0]

  387. 	fmla	v23.4s, v6.4s, v8.s[1]

  388. 	fmla	v27.4s, v6.4s, v8.s[2]

  389. 	fmla	v31.4s, v6.4s, v8.s[3]

  390. .endm

  391. 

  392. .macro SAVE16x4

  393. 	mov	pCRow1, pCRow0

  394. 

  395. 	ld1	{v0.4s, v1.4s, v2.4s, v3.4s}, [pCRow1]

  396. 	fmla	v0.4s, v16.4s, alphaV0

  397. 	fmla	v1.4s, v17.4s, alphaV1

  398. 	fmla	v2.4s, v18.4s, alphaV2

  399. 	fmla	v3.4s, v19.4s, alphaV3

  400. 	st1 	{v0.4s, v1.4s, v2.4s, v3.4s}, [pCRow1]

  401. 

  402. 	add	pCRow1, pCRow1, LDC

  403. 

  404. 	ld1	{v4.4s, v5.4s, v6.4s, v7.4s}, [pCRow1]

  405. 	fmla	v4.4s, v20.4s, alphaV0

  406. 	fmla	v5.4s, v21.4s, alphaV1

  407. 	fmla	v6.4s, v22.4s, alphaV2

  408. 	fmla	v7.4s, v23.4s, alphaV3

  409. 	st1 	{v4.4s, v5.4s, v6.4s, v7.4s}, [pCRow1]

  410. 

  411. 	add	pCRow1, pCRow1, LDC

  412. 

  413. 	ld1	{v0.4s, v1.4s, v2.4s, v3.4s}, [pCRow1]

  414. 	fmla	v0.4s, v24.4s, alphaV0

  415. 	fmla	v1.4s, v25.4s, alphaV1

  416. 	fmla	v2.4s, v26.4s, alphaV2

  417. 	fmla	v3.4s, v27.4s, alphaV3

  418. 	st1 	{v0.4s, v1.4s, v2.4s, v3.4s}, [pCRow1]

  419. 

  420. 	add	pCRow1, pCRow1, LDC

  421. 

  422. 	ld1	{v4.4s, v5.4s, v6.4s, v7.4s}, [pCRow1]

  423. 	fmla	v4.4s, v28.4s, alphaV0

  424. 	fmla	v5.4s, v29.4s, alphaV1

  425. 	fmla	v6.4s, v30.4s, alphaV2

  426. 	fmla	v7.4s, v31.4s, alphaV3

  427. 	st1 	{v4.4s, v5.4s, v6.4s, v7.4s}, [pCRow1]

  428. 

  429. 	add	pCRow0, pCRow0, #64

  430. .endm

  431. 

  432. /******************************************************************************/

  433. 

  434. .macro INIT8x4

  435. 	fmov		s16, wzr

  436. 	fmov		s17, s16

  437. 	fmov		s18, s17

  438. 	fmov		s19, s16

  439. 	fmov		s20, s17

  440. 	fmov		s21, s16

  441. 	fmov		s22, s17

  442. 	fmov		s23, s16

  443. 	fmov		s24, s17

  444. 	fmov		s25, s16

  445. 	fmov		s26, s17

  446. 	fmov		s27, s16

  447. 	fmov		s28, s17

  448. 	fmov		s29, s16

  449. 	fmov		s30, s17

  450. 	fmov		s31, s16

  451. .endm

  452. 

  453. .macro KERNEL8x4_SUB

  454. 	ld1	{v8.2s, v9.2s}, [pB]

  455. 	add	pB, pB, #16

  456. 	ld1	{v0.2s, v1.2s}, [pA_0]

  457. 	add	pA_0, pA_0, #16

  458. 

  459. 	fmla	v16.2s, v0.2s, v8.s[0]

  460. 	fmla	v29.2s, v1.2s, v9.s[1]

  461. 	fmla	v20.2s, v0.2s, v8.s[1]

  462. 	fmla	v25.2s, v1.2s, v9.s[0]

  463. 

  464. 	ld1	{v2.2s, v3.2s}, [pA_1]

  465. 	add	pA_1, pA_1, #16

  466. 

  467. 	fmla	v24.2s, v0.2s, v9.s[0]

  468. 	fmla	v21.2s, v1.2s, v8.s[1]

  469. 	fmla	v28.2s, v0.2s, v9.s[1]

  470. 	fmla	v17.2s, v1.2s, v8.s[0]

  471. 

  472. 	fmla	v18.2s, v2.2s, v8.s[0]

  473. 	fmla	v31.2s, v3.2s, v9.s[1]

  474. 	fmla	v22.2s, v2.2s, v8.s[1]

  475. 	fmla	v27.2s, v3.2s, v9.s[0]

  476. 

  477. 	fmla	v26.2s, v2.2s, v9.s[0]

  478. 	fmla	v23.2s, v3.2s, v8.s[1]

  479. 	fmla	v30.2s, v2.2s, v9.s[1]

  480. 	fmla	v19.2s, v3.2s, v8.s[0]

  481. .endm

  482. 

  483. .macro SAVE8x4

  484. 	mov	pCRow1, pCRow0

  485. 

  486. 	ld1	{v0.2s, v1.2s}, [pCRow1]

  487. 	fmla	v0.2s, v16.2s, alphaV0

  488. 	fmla	v1.2s, v17.2s, alphaV1

  489. 	st1 	{v0.2s, v1.2s}, [pCRow1]

  490. 

  491. 	add	pCRow2, pCRow1, LDC

  492. 	add	pCRow1, pCRow1, #16

  493. 

  494. 	ld1	{v2.2s, v3.2s}, [pCRow1]

  495. 	fmla	v2.2s, v18.2s, alphaV2

  496. 	fmla	v3.2s, v19.2s, alphaV3

  497. 	st1 	{v2.2s, v3.2s}, [pCRow1]

  498. 

  499. 	ld1	{v4.2s, v5.2s}, [pCRow2]

  500. 	fmla	v4.2s, v20.2s, alphaV0

  501. 	fmla	v5.2s, v21.2s, alphaV1

  502. 	st1 	{v4.2s, v5.2s}, [pCRow2]

  503. 

  504. 	add	pCRow1, pCRow2, LDC

  505. 	add	pCRow2, pCRow2, #16

  506. 

  507. 	ld1	{v6.2s, v7.2s}, [pCRow2]

  508. 	fmla	v6.2s, v22.2s, alphaV2

  509. 	fmla	v7.2s, v23.2s, alphaV3

  510. 	st1 	{v6.2s, v7.2s}, [pCRow2]

  511. 

  512. 	ld1	{v0.2s, v1.2s}, [pCRow1]

  513. 	fmla	v0.2s, v24.2s, alphaV0

  514. 	fmla	v1.2s, v25.2s, alphaV1

  515. 	st1 	{v0.2s, v1.2s}, [pCRow1]

  516. 

  517. 	add	pCRow2, pCRow1, LDC

  518. 	add	pCRow1, pCRow1, #16

  519. 

  520. 	ld1	{v2.2s, v3.2s}, [pCRow1]

  521. 	fmla	v2.2s, v26.2s, alphaV2

  522. 	fmla	v3.2s, v27.2s, alphaV3

  523. 	st1 	{v2.2s, v3.2s}, [pCRow1]

  524. 

  525. 	ld1	{v4.2s, v5.2s}, [pCRow2]

  526. 	fmla	v4.2s, v28.2s, alphaV0

  527. 	fmla	v5.2s, v29.2s, alphaV1

  528. 	st1 	{v4.2s, v5.2s}, [pCRow2]

  529. 

  530. 	add	pCRow2, pCRow2, #16

  531. 

  532. 	ld1	{v6.2s, v7.2s}, [pCRow2]

  533. 	fmla	v6.2s, v30.2s, alphaV2

  534. 	fmla	v7.2s, v31.2s, alphaV3

  535. 	st1 	{v6.2s, v7.2s}, [pCRow2]

  536. 

  537. 	add	pCRow0, pCRow0, #32

  538. .endm

  539. 

  540. /******************************************************************************/

  541. 

  542. .macro INIT4x4

  543. 	fmov		s16, wzr

  544. 	fmov		s17, s16

  545. 	fmov		s20, s17

  546. 	fmov		s21, s16

  547. 	fmov		s24, s17

  548. 	fmov		s25, s16

  549. 	fmov		s28, s17

  550. 	fmov		s29, s16

  551. .endm

  552. 

  553. .macro KERNEL4x4_SUB

  554. 	ld1	{v8.2s, v9.2s}, [pB]

  555. 	add	pB, pB, #16

  556. 	ld1	{v0.2s, v1.2s}, [pA_0]

  557. 	add	pA_0, pA_0, #16

  558. 

  559. 	fmla	v16.2s, v0.2s, v8.s[0]

  560. 	fmla	v29.2s, v1.2s, v9.s[1]

  561. 

  562. 	fmla	v20.2s, v0.2s, v8.s[1]

  563. 	fmla	v25.2s, v1.2s, v9.s[0]

  564. 

  565. 	fmla	v24.2s, v0.2s, v9.s[0]

  566. 	fmla	v21.2s, v1.2s, v8.s[1]

  567. 

  568. 	fmla	v28.2s, v0.2s, v9.s[1]

  569. 	fmla	v17.2s, v1.2s, v8.s[0]

  570. .endm

  571. 

  572. .macro SAVE4x4

  573. 	ld1	{v8.2s, v9.2s}, [pCRow0]

  574. 	fmla	v8.2s, v16.2s, alphaV0

  575. 	fmla	v9.2s, v17.2s, alphaV1

  576. 	st1 	{v8.2s, v9.2s}, [pCRow0]

  577. 

  578. 	add	pCRow1, pCRow0, LDC

  579. 

  580. 	ld1	{v12.2s, v13.2s}, [pCRow1]

  581. 	fmla	v12.2s, v20.2s, alphaV2

  582. 	fmla	v13.2s, v21.2s, alphaV3

  583. 	st1 	{v12.2s, v13.2s}, [pCRow1]

  584. 

  585. 	add	pCRow2, pCRow1, LDC

  586. 

  587. 	ld1	{v8.2s, v9.2s}, [pCRow2]

  588. 	fmla	v8.2s, v24.2s, alphaV0

  589. 	fmla	v9.2s, v25.2s, alphaV1

  590. 	st1 	{v8.2s, v9.2s}, [pCRow2]

  591. 

  592. 	add	pCRow1, pCRow2, LDC

  593. 

  594. 	ld1	{v12.2s, v13.2s}, [pCRow1]

  595. 	fmla	v12.2s, v28.2s, alphaV2

  596. 	fmla	v13.2s, v29.2s, alphaV3

  597. 	st1 	{v12.2s, v13.2s}, [pCRow1]

  598. 

  599. 	add	pCRow0, pCRow0, #16

  600. .endm

  601. 

  602. /******************************************************************************/

  603. 

  604. .macro INIT2x4

  605. 	fmov		s16, wzr

  606. 	fmov		s20, s16

  607. 	fmov		s24, s20

  608. 	fmov		s28, s16

  609. .endm

  610. 

  611. .macro KERNEL2x4_SUB

  612. 	ld1	{v8.2s, v9.2s}, [pB]

  613. 	add	pB, pB, #16

  614. 	ld1	{v0.2s}, [pA_0]

  615. 	add	pA_0, pA_0, #8

  616. 

  617. 	fmla	v16.2s, v0.2s, v8.s[0]

  618. 	fmla	v20.2s, v0.2s, v8.s[1]

  619. 	fmla	v24.2s, v0.2s, v9.s[0]

  620. 	fmla	v28.2s, v0.2s, v9.s[1]

  621. .endm

  622. 

  623. .macro SAVE2x4

  624. 	ld1	{v8.2s}, [pCRow0]

  625. 	fmla	v8.2s, v16.2s, alphaV0

  626. 	st1	{v8.2s}, [pCRow0]

  627. 

  628. 	add	pCRow1, pCRow0, LDC

  629. 

  630. 	ld1	{v12.2s}, [pCRow1]

  631. 	fmla	v12.2s, v20.2s, alphaV1

  632. 	st1	{v12.2s}, [pCRow1]

  633. 

  634. 	add	pCRow2, pCRow1, LDC

  635. 

  636. 	ld1	{v8.2s}, [pCRow2]

  637. 	fmla	v8.2s, v24.2s, alphaV2

  638. 	st1	{v8.2s}, [pCRow2]

  639. 

  640. 	add	pCRow1, pCRow2, LDC

  641. 

  642. 	ld1	{v12.2s}, [pCRow1]

  643. 	fmla	v12.2s, v28.2s, alphaV3

  644. 	st1	{v12.2s}, [pCRow1]

  645. 

  646. 	add	pCRow0, pCRow0, #8

  647. .endm

  648. 

  649. /******************************************************************************/

  650. 

  651. .macro INIT1x4

  652. 	fmov		s16, wzr

  653. 	fmov		s20, s16

  654. .endm

  655. 

  656. .macro KERNEL1x4_SUB

  657. 	ldr	s0, [pA_0]

  658. 	add	pA_0, pA_0, #4

  659. 

  660. 	ld1	{v8.2s, v9.2s}, [pB]

  661. 	add	pB, pB, #16

  662. 

  663. 	fmla	v16.2s, v8.2s, v0.s[0]

  664. 	fmla	v20.2s, v9.2s, v0.s[0]

  665. .endm

  666. 

  667. .macro SAVE1x4

  668. 	add	pCRow1, pCRow0, LDC

  669. 

  670. 	ld1	{v8.s}[0], [pCRow0]

  671. 	ld1	{v8.s}[1], [pCRow1]

  672. 	fmla	v8.2s, v16.2s, alphaV0

  673. 	st1	{v8.s}[0], [pCRow0]

  674. 	st1	{v8.s}[1], [pCRow1]

  675. 

  676. 	add	pCRow2, pCRow1, LDC

  677. 	add	pCRow1, pCRow2, LDC

  678. 

  679. 	ld1	{v12.s}[0], [pCRow2]

  680. 	ld1	{v12.s}[1], [pCRow1]

  681. 	fmla	v12.2s, v20.2s, alphaV1

  682. 	st1	{v12.s}[0], [pCRow2]

  683. 	st1	{v12.s}[1], [pCRow1]

  684. 

  685. 	add	pCRow0, pCRow0, #4

  686. .endm

  687. 

  688. /******************************************************************************/

  689. 

  690. .macro INIT4x2

  691. 	fmov	s16, wzr

  692. 	fmov	s17, s16

  693. 	fmov	s20, s17

  694. 	fmov	s21, s16

  695. .endm

  696. 

  697. .macro KERNEL4x2_SUB

  698. 	ld1	{v8.2s}, [pB]

  699. 	add	pB, pB, #8

  700. 	ld1	{v0.2s, v1.2s}, [pA_0]

  701. 	add	pA_0, pA_0, #16

  702. 

  703. 	fmla	v16.2s, v0.2s, v8.s[0]

  704. 	fmla	v17.2s, v1.2s, v8.s[0]

  705. 	fmla	v20.2s, v0.2s, v8.s[1]

  706. 	fmla	v21.2s, v1.2s, v8.s[1]

  707. .endm

  708. 

  709. .macro SAVE4x2

  710. 	ld1	{v8.2s, v9.2s}, [pCRow0]

  711. 	fmla	v8.2s, v16.2s, alphaV0

  712. 	fmla	v9.2s, v17.2s, alphaV1

  713. 	st1	{v8.2s, v9.2s}, [pCRow0]

  714. 

  715. 	add	pCRow1, pCRow0, LDC

  716. 

  717. 	ld1	{v12.2s, v13.2s}, [pCRow1]

  718. 	fmla	v12.2s, v20.2s, alphaV2

  719. 	fmla	v13.2s, v21.2s, alphaV3

  720. 	st1	{v12.2s, v13.2s}, [pCRow1]

  721. 

  722. 	add	pCRow0, pCRow0, #16

  723. .endm

  724. 

  725. /******************************************************************************/

  726. 

  727. .macro INIT2x2

  728. 	fmov		s16, wzr

  729. 	fmov		s20, s16

  730. .endm

  731. 

  732. .macro KERNEL2x2_SUB

  733. 	ld1	{v8.2s}, [pB]

  734. 	add	pB, pB, #8

  735. 

  736. 	ld1	{v0.2s}, [pA_0]

  737. 	add	pA_0, pA_0, #8

  738. 

  739. 	fmla	v16.2s, v0.2s, v8.s[0]

  740. 	fmla	v20.2s, v0.2s, v8.s[1]

  741. .endm

  742. 

  743. .macro SAVE2x2

  744. 	ld1	{v8.2s}, [pCRow0]

  745. 	fmla	v8.2s, v16.2s, alphaV0

  746. 	st1	{v8.2s}, [pCRow0]

  747. 

  748. 	add	pCRow1 , pCRow0, LDC

  749. 

  750. 	ld1	{v12.2s}, [pCRow1]

  751. 	fmla	v12.2s, v20.2s, alphaV1

  752. 	st1	{v12.2s}, [pCRow1]

  753. 

  754. 	add	pCRow0, pCRow0, #8

  755. .endm

  756. 

  757. /******************************************************************************/

  758. 

  759. .macro INIT1x2

  760. 	fmov		s16, wzr

  761. .endm

  762. 

  763. .macro KERNEL1x2_SUB

  764. 	ld1	{v8.2s} , [pB]

  765. 	add	pB , pB, #8

  766. 

  767. 	ldr	s0 , [pA_0]

  768. 	add	pA_0, pA_0, #4

  769. 

  770. 	fmla	v16.2s, v8.2s, v0.s[0]

  771. .endm

  772. 

  773. .macro SAVE1x2

  774. 	add	pCRow1 , pCRow0, LDC

  775. 

  776. 	ld1	{v8.s}[0], [pCRow0]

  777. 	ld1	{v8.s}[1], [pCRow1]

  778. 	fmla	v8.2s, v16.2s, alphaV0

  779. 	st1	{v8.s}[0], [pCRow0]

  780. 	st1	{v8.s}[1], [pCRow1]

  781. 

  782. 	add	pCRow0, pCRow0, #4

  783. .endm

  784. 

  785. /******************************************************************************/

  786. 

  787. .macro INIT4x1

  788. 	fmov	s16, wzr

  789. 	fmov	s17, s16

  790. .endm

  791. 

  792. .macro KERNEL4x1_SUB

  793. 	ldr	s8, [pB]

  794. 	add	pB , pB, #4

  795. 

  796. 	ld1	{v0.2s, v1.2s}, [pA_0]

  797. 	add	pA_0 , pA_0, #16

  798. 

  799. 	fmla	v16.2s, v0.2s, v8.s[0]

  800. 	fmla	v17.2s, v1.2s, v8.s[0]

  801. .endm

  802. 

  803. .macro SAVE4x1

  804. 	ld1	{v8.2s, v9.2s}, [pCRow0]

  805. 	fmla	v8.2s, v16.2s, alphaV0

  806. 	fmla	v9.2s, v17.2s, alphaV1

  807. 	st1	{v8.2s, v9.2s}, [pCRow0]

  808. 

  809. 	add	pCRow0, pCRow0, #16

  810. .endm

  811. 

  812. 

  813. 

  814. 

  815. /******************************************************************************/

  816. 

  817. .macro INIT2x1

  818. 	fmov		s16, wzr

  819. .endm

  820. 

  821. .macro KERNEL2x1_SUB

  822. 	ldr	s8, [pB]

  823. 	add	pB , pB, #4

  824. 

  825. 	ld1	{v0.2s}, [pA_0]

  826. 	add	pA_0 , pA_0, #8

  827. 

  828. 	fmla	v16.2s, v0.2s, v8.s[0]

  829. .endm

  830. 

  831. .macro SAVE2x1

  832. 	ld1	{v8.2s}, [pCRow0]

  833. 	fmla	v8.2s, v16.2s, alphaV0

  834. 	st1	{v8.2s}, [pCRow0]

  835. 

  836. 	add	pCRow0, pCRow0, #8

  837. .endm

  838. 

  839. /******************************************************************************/

  840. 

  841. .macro INIT1x1

  842. 	fmov	s16, wzr

  843. .endm

  844. 

  845. .macro KERNEL1x1_SUB

  846. 	ldr	s8, [pB]

  847. 	add	pB , pB, #4

  848. 

  849. 	ldr	s0, [pA_0]

  850. 	add	pA_0 , pA_0, #4

  851. 

  852. 	fmadd 	s16, s0, s8, s16  

  853. .endm

  854. 

  855. .macro SAVE1x1

  856. 	ldr	s8, [pCRow0]

  857. 	fmadd	s8, s16, alpha0, s8

  858. 	str 	s8, [pCRow0]

  859. 

  860. 	add	pCRow0, pCRow0, #4

  861. .endm

  862. 

  863. /*******************************************************************************

  864. * End of macro definitions

  865. *******************************************************************************/

  866. 

  867. 	PROLOGUE

  868. 

  869. 	.align 5

  870. 	add	sp, sp, #-(11 * 16)

  871. 	stp	d8, d9, [sp, #(0 * 16)]

  872. 	stp	d10, d11, [sp, #(1 * 16)]

  873. 	stp	d12, d13, [sp, #(2 * 16)]

  874. 	stp	d14, d15, [sp, #(3 * 16)]

  875. 	stp	d16, d17, [sp, #(4 * 16)]

  876. 	stp	x18, x19, [sp, #(5 * 16)]

  877. 	stp	x20, x21, [sp, #(6 * 16)]

  878. 	stp	x22, x23, [sp, #(7 * 16)]

  879. 	stp	x24, x25, [sp, #(8 * 16)]

  880. 	stp	x26, x27, [sp, #(9 * 16)]

  881. 	str	x28, [sp, #(10 * 16)]

  882. 

  883. 	fmov	alpha0, s0

  884. 	fmov	alpha1, s0

  885. 	fmov	alpha2, s0

  886. 	fmov	alpha3, s0

  887. 

  888. 	lsl	LDC, LDC, #2			// ldc = ldc * 4

  889. 

  890. 	mov	pB, origPB

  891. 

  892. 	mov	counterJ, origN

  893. 	asr 	counterJ, counterJ, #2		// J = J / 4

  894. 	cmp 	counterJ, #0

  895. 	ble	.Lsgemm_kernel_L2_BEGIN

  896. 

  897. /******************************************************************************/

  898. 

  899. .Lsgemm_kernel_L4_BEGIN:

  900. 	mov	pCRow0, pC			// pCRow0 = C

  901. 	add	pC, pC, LDC, lsl #2

  902. 

  903. 	lsl	temp, origK, #4			// k * 4 * 4

  904. 	mov	pA_0, origPA			// pA_0 = start of A array

  905. 	add	pA_1, temp, pA_0

  906. 	add	pA_2, temp, pA_1

  907. 	add	pA_3, temp, pA_2

  908. 

  909. .Lsgemm_kernel_L4_M16_BEGIN:

  910. 

  911. 	mov	counterI, origM

  912. 	asr 	counterI, counterI, #4		// counterI = counterI / 16

  913. 	cmp 	counterI, #0

  914. 	ble	.Lsgemm_kernel_L4_M8_BEGIN

  915. 

  916. .Lsgemm_kernel_L4_M16_20:

  917. 

  918. 	mov	pB, origPB

  919. 	asr 	counterL , origK, #1		// L = K / 2

  920. 	cmp	counterL , #2			// is there at least 4 to do?

  921. 	blt	.Lsgemm_kernel_L4_M16_32

  922. 

  923. 	KERNEL16x4_I				// do one in the K

  924. 	KERNEL16x4_M2				// do another in the K

  925. 

  926. 	subs	counterL, counterL, #2

  927. 	ble	.Lsgemm_kernel_L4_M16_22a

  928. 	.align 5

  929. 

  930. .Lsgemm_kernel_L4_M16_22:

  931. 

  932. 	KERNEL16x4_M1

  933. 	KERNEL16x4_M2

  934. 

  935. 	subs	counterL, counterL, #1

  936. 	bgt	.Lsgemm_kernel_L4_M16_22

  937. 

  938. 

  939. .Lsgemm_kernel_L4_M16_22a:

  940. 

  941. 	KERNEL16x4_M1

  942. 	KERNEL16x4_E

  943. 

  944. 	b	 .Lsgemm_kernel_L4_M16_44

  945. 

  946. .Lsgemm_kernel_L4_M16_32:

  947. 

  948. 	tst	counterL, #1

  949. 	ble	.Lsgemm_kernel_L4_M16_40

  950. 

  951. 	KERNEL16x4_I

  952. 

  953. 	KERNEL16x4_E

  954. 

  955. 	b	.Lsgemm_kernel_L4_M16_44

  956. 

  957. 

  958. .Lsgemm_kernel_L4_M16_40:

  959. 

  960. 	INIT16x4

  961. 

  962. .Lsgemm_kernel_L4_M16_44:

  963. 

  964. 	ands	counterL , origK, #1

  965. 	ble	.Lsgemm_kernel_L4_M16_100

  966. 

  967. .Lsgemm_kernel_L4_M16_46:

  968. 

  969. 	KERNEL16x4_SUB

  970. 

  971. .Lsgemm_kernel_L4_M16_100:

  972. 

  973. 	SAVE16x4

  974. 

  975. .Lsgemm_kernel_L4_M16_END:

  976. 	lsl	temp, origK, #4			// k * 4 * 4 = Four rows of A

  977. 	add	pA_0, pA_0, temp

  978. 	add	pA_0, pA_0, temp

  979. 	add	pA_0, pA_0, temp

  980. 	add	pA_1, pA_0, temp

  981. 	add	pA_2, pA_1, temp

  982. 	add	pA_3, pA_2, temp

  983. 	subs	counterI, counterI, #1

  984. 	bne	.Lsgemm_kernel_L4_M16_20

  985. 

  986. .Lsgemm_kernel_L4_M8_BEGIN:

  987. 	mov	counterI, origM

  988. 	tst	counterI , #15

  989. 	ble	.Lsgemm_kernel_L4_END

  990. 

  991. 	tst	counterI, #8

  992. 	ble	.Lsgemm_kernel_L4_M4_BEGIN

  993. 

  994. .Lsgemm_kernel_L4_M8_20:

  995. 

  996. 	INIT8x4

  997. 

  998. 	mov	pB, origPB

  999. 	asr 	counterL, origK, #3		// counterL = counterL / 8

  1000. 	cmp	counterL, #0

  1001. 	ble	.Lsgemm_kernel_L4_M8_40

  1002. 

  1003. .Lsgemm_kernel_L4_M8_22:

  1004. 

  1005. 	KERNEL8x4_SUB

  1006. 	KERNEL8x4_SUB

  1007. 	KERNEL8x4_SUB

  1008. 	KERNEL8x4_SUB

  1009. 

  1010. 	KERNEL8x4_SUB

  1011. 	KERNEL8x4_SUB

  1012. 	KERNEL8x4_SUB

  1013. 	KERNEL8x4_SUB

  1014. 

  1015. 	subs	counterL, counterL, #1

  1016. 	bgt	.Lsgemm_kernel_L4_M8_22

  1017. 

  1018. 

  1019. .Lsgemm_kernel_L4_M8_40:

  1020. 

  1021. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1022. 	ble	.Lsgemm_kernel_L4_M8_100

  1023. 

  1024. .Lsgemm_kernel_L4_M8_42:

  1025. 

  1026. 	KERNEL8x4_SUB

  1027. 

  1028. 	subs	counterL, counterL, #1

  1029. 	bgt	.Lsgemm_kernel_L4_M8_42

  1030. 

  1031. .Lsgemm_kernel_L4_M8_100:

  1032. 

  1033. 	SAVE8x4

  1034. 

  1035. .Lsgemm_kernel_L4_M8_END:

  1036. 	lsl	temp, origK, #4			// k * 4 * 4

  1037. 	add	pA_0, pA_0, temp

  1038. 

  1039. .Lsgemm_kernel_L4_M4_BEGIN:

  1040. 	mov	counterI, origM

  1041. 	tst	counterI , #7

  1042. 	ble	.Lsgemm_kernel_L4_END

  1043. 

  1044. 	tst	counterI, #4

  1045. 	ble	.Lsgemm_kernel_L4_M2_BEGIN

  1046. 

  1047. .Lsgemm_kernel_L4_M4_20:

  1048. 

  1049. 	INIT4x4

  1050. 

  1051. 	mov	pB, origPB

  1052. 	asr 	counterL, origK, #3		// counterL = counterL / 8

  1053. 	cmp	counterL, #0

  1054. 	ble	.Lsgemm_kernel_L4_M4_40

  1055. 

  1056. .Lsgemm_kernel_L4_M4_22:

  1057. 

  1058. 	KERNEL4x4_SUB

  1059. 	KERNEL4x4_SUB

  1060. 	KERNEL4x4_SUB

  1061. 	KERNEL4x4_SUB

  1062. 

  1063. 	KERNEL4x4_SUB

  1064. 	KERNEL4x4_SUB

  1065. 	KERNEL4x4_SUB

  1066. 	KERNEL4x4_SUB

  1067. 

  1068. 	subs	counterL, counterL, #1

  1069. 	bgt	.Lsgemm_kernel_L4_M4_22

  1070. 

  1071. 

  1072. .Lsgemm_kernel_L4_M4_40:

  1073. 

  1074. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1075. 	ble	.Lsgemm_kernel_L4_M4_100

  1076. 

  1077. .Lsgemm_kernel_L4_M4_42:

  1078. 

  1079. 	KERNEL4x4_SUB

  1080. 

  1081. 	subs	counterL, counterL, #1

  1082. 	bgt	.Lsgemm_kernel_L4_M4_42

  1083. 

  1084. .Lsgemm_kernel_L4_M4_100:

  1085. 

  1086. 	SAVE4x4

  1087. 

  1088. .Lsgemm_kernel_L4_M4_END:

  1089. 

  1090. 

  1091. .Lsgemm_kernel_L4_M2_BEGIN:

  1092. 

  1093. 	mov	counterI, origM

  1094. 	tst	counterI , #3

  1095. 	ble	.Lsgemm_kernel_L4_END

  1096. 

  1097. 	tst	counterI, #2			// counterI = counterI / 2

  1098. 	ble	.Lsgemm_kernel_L4_M1_BEGIN

  1099. 

  1100. .Lsgemm_kernel_L4_M2_20:

  1101. 

  1102. 	INIT2x4

  1103. 

  1104. 	mov	pB, origPB

  1105. 	asr 	counterL , origK, #3		// counterL = counterL / 8

  1106. 	cmp	counterL , #0

  1107. 	ble	.Lsgemm_kernel_L4_M2_40

  1108. 

  1109. .Lsgemm_kernel_L4_M2_22:

  1110. 

  1111. 	KERNEL2x4_SUB

  1112. 	KERNEL2x4_SUB

  1113. 	KERNEL2x4_SUB

  1114. 	KERNEL2x4_SUB

  1115. 

  1116. 	KERNEL2x4_SUB

  1117. 	KERNEL2x4_SUB

  1118. 	KERNEL2x4_SUB

  1119. 	KERNEL2x4_SUB

  1120. 

  1121. 	subs	counterL, counterL, #1

  1122. 	bgt	.Lsgemm_kernel_L4_M2_22

  1123. 

  1124. 

  1125. .Lsgemm_kernel_L4_M2_40:

  1126. 

  1127. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1128. 	ble	.Lsgemm_kernel_L4_M2_100

  1129. 

  1130. .Lsgemm_kernel_L4_M2_42:

  1131. 

  1132. 	KERNEL2x4_SUB

  1133. 

  1134. 	subs	counterL, counterL, #1

  1135. 	bgt	.Lsgemm_kernel_L4_M2_42

  1136. 

  1137. .Lsgemm_kernel_L4_M2_100:

  1138. 

  1139. 	SAVE2x4

  1140. 

  1141. .Lsgemm_kernel_L4_M2_END:

  1142. 

  1143. 

  1144. .Lsgemm_kernel_L4_M1_BEGIN:

  1145. 

  1146. 	tst	counterI, #1			// counterI = counterI % 2

  1147. 	ble	.Lsgemm_kernel_L4_END

  1148. 

  1149. .Lsgemm_kernel_L4_M1_20:

  1150. 

  1151. 	INIT1x4

  1152. 

  1153. 	mov	pB, origPB

  1154. 	asr 	counterL , origK, #3		// counterL = counterL / 8

  1155. 	cmp	counterL , #0

  1156. 	ble	.Lsgemm_kernel_L4_M1_40

  1157. 

  1158. .Lsgemm_kernel_L4_M1_22:

  1159. 	KERNEL1x4_SUB

  1160. 	KERNEL1x4_SUB

  1161. 	KERNEL1x4_SUB

  1162. 	KERNEL1x4_SUB

  1163. 

  1164. 	KERNEL1x4_SUB

  1165. 	KERNEL1x4_SUB

  1166. 	KERNEL1x4_SUB

  1167. 	KERNEL1x4_SUB

  1168. 

  1169. 	subs	counterL, counterL, #1

  1170. 	bgt	.Lsgemm_kernel_L4_M1_22

  1171. 

  1172. 

  1173. .Lsgemm_kernel_L4_M1_40:

  1174. 

  1175. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1176. 	ble	.Lsgemm_kernel_L4_M1_100

  1177. 

  1178. .Lsgemm_kernel_L4_M1_42:

  1179. 

  1180. 	KERNEL1x4_SUB

  1181. 

  1182. 	subs	counterL, counterL, #1

  1183. 	bgt	.Lsgemm_kernel_L4_M1_42

  1184. 

  1185. .Lsgemm_kernel_L4_M1_100:

  1186. 

  1187. 	SAVE1x4

  1188. 

  1189. 

  1190. .Lsgemm_kernel_L4_END:

  1191. 

  1192. 	lsl	temp, origK, #4 

  1193. 	add	origPB, origPB, temp		// B = B + K * 4 * 4

  1194. 

  1195. 	subs	counterJ, counterJ , #1		// j--

  1196. 	bgt	.Lsgemm_kernel_L4_BEGIN

  1197. 

  1198. 

  1199. /******************************************************************************/

  1200. 

  1201. .Lsgemm_kernel_L2_BEGIN:   // less than 2 left in N direction

  1202. 

  1203. 	mov	counterJ , origN

  1204. 	tst	counterJ , #3

  1205. 	ble	.Lsgemm_kernel_L999

  1206. 

  1207. 	tst	counterJ , #2

  1208. 	ble	.Lsgemm_kernel_L1_BEGIN

  1209. 

  1210. 	mov	pCRow0, pC			// pCRow0 = pC

  1211. 

  1212. 	add	pC,pC,LDC, lsl #1

  1213. 

  1214. 	mov	pA_0, origPA			// pA_0 = A

  1215. 

  1216. 

  1217. 

  1218. .Lsgemm_kernel_L2_M4_BEGIN:

  1219. 

  1220. 	mov	counterI, origM

  1221. 	asr 	counterI, counterI, #2		// counterI = counterI / 4

  1222. 	cmp	counterI,#0

  1223. 	ble	.Lsgemm_kernel_L2_M2_BEGIN

  1224. 

  1225. .Lsgemm_kernel_L2_M4_20:

  1226. 

  1227. 	INIT4x2

  1228. 

  1229. 	mov	pB, origPB

  1230. 	asr	counterL , origK, #3		// counterL = counterL / 8

  1231. 	cmp	counterL,#0

  1232. 	ble	.Lsgemm_kernel_L2_M4_40

  1233. 	.align 5

  1234. 

  1235. .Lsgemm_kernel_L2_M4_22:

  1236. 	KERNEL4x2_SUB

  1237. 	KERNEL4x2_SUB

  1238. 	KERNEL4x2_SUB

  1239. 	KERNEL4x2_SUB

  1240. 

  1241. 	KERNEL4x2_SUB

  1242. 	KERNEL4x2_SUB

  1243. 	KERNEL4x2_SUB

  1244. 	KERNEL4x2_SUB

  1245. 

  1246. 	subs	counterL, counterL, #1

  1247. 	bgt	.Lsgemm_kernel_L2_M4_22

  1248. 

  1249. 

  1250. .Lsgemm_kernel_L2_M4_40:

  1251. 

  1252. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1253. 	ble	.Lsgemm_kernel_L2_M4_100

  1254. 

  1255. .Lsgemm_kernel_L2_M4_42:

  1256. 

  1257. 	KERNEL4x2_SUB

  1258. 

  1259. 	subs	counterL, counterL, #1

  1260. 	bgt	.Lsgemm_kernel_L2_M4_42

  1261. 

  1262. .Lsgemm_kernel_L2_M4_100:

  1263. 

  1264. 	SAVE4x2

  1265. 

  1266. .Lsgemm_kernel_L2_M4_END:

  1267. 

  1268. 	subs	counterI, counterI, #1

  1269. 	bgt	.Lsgemm_kernel_L2_M4_20

  1270. 

  1271. 

  1272. .Lsgemm_kernel_L2_M2_BEGIN:

  1273. 

  1274. 	mov	counterI, origM

  1275. 	tst	counterI , #3

  1276. 	ble	.Lsgemm_kernel_L2_END

  1277. 

  1278. 	tst	counterI, #2			// counterI = counterI / 2

  1279. 	ble	.Lsgemm_kernel_L2_M1_BEGIN

  1280. 

  1281. .Lsgemm_kernel_L2_M2_20:

  1282. 

  1283. 	INIT2x2

  1284. 

  1285. 	mov	pB, origPB

  1286. 	asr	counterL , origK, #3		// counterL = counterL / 8

  1287.         cmp	counterL,#0

  1288. 	ble	.Lsgemm_kernel_L2_M2_40

  1289. 

  1290. .Lsgemm_kernel_L2_M2_22:

  1291. 

  1292. 	KERNEL2x2_SUB

  1293. 	KERNEL2x2_SUB

  1294. 	KERNEL2x2_SUB

  1295. 	KERNEL2x2_SUB

  1296. 

  1297. 	KERNEL2x2_SUB

  1298. 	KERNEL2x2_SUB

  1299. 	KERNEL2x2_SUB

  1300. 	KERNEL2x2_SUB

  1301. 

  1302. 	subs	counterL, counterL, #1

  1303. 	bgt	.Lsgemm_kernel_L2_M2_22

  1304. 

  1305. 

  1306. .Lsgemm_kernel_L2_M2_40:

  1307. 

  1308. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1309. 	ble	.Lsgemm_kernel_L2_M2_100

  1310. 

  1311. .Lsgemm_kernel_L2_M2_42:

  1312. 

  1313. 	KERNEL2x2_SUB

  1314. 

  1315. 	subs	counterL, counterL, #1

  1316. 	bgt	.Lsgemm_kernel_L2_M2_42

  1317. 

  1318. .Lsgemm_kernel_L2_M2_100:

  1319. 

  1320. 	SAVE2x2

  1321. 

  1322. .Lsgemm_kernel_L2_M2_END:

  1323. 

  1324. 

  1325. .Lsgemm_kernel_L2_M1_BEGIN:

  1326. 

  1327. 	tst	counterI, #1			// counterI = counterI % 2

  1328. 	ble	.Lsgemm_kernel_L2_END

  1329. 

  1330. .Lsgemm_kernel_L2_M1_20:

  1331. 

  1332. 	INIT1x2

  1333. 

  1334. 	mov	pB, origPB

  1335. 	asr 	counterL , origK, #3		// counterL = counterL / 8

  1336.         cmp     counterL, #0

  1337. 	ble	.Lsgemm_kernel_L2_M1_40

  1338. 

  1339. .Lsgemm_kernel_L2_M1_22:

  1340. 	KERNEL1x2_SUB

  1341. 	KERNEL1x2_SUB

  1342. 	KERNEL1x2_SUB

  1343. 	KERNEL1x2_SUB

  1344. 

  1345. 	KERNEL1x2_SUB

  1346. 	KERNEL1x2_SUB

  1347. 	KERNEL1x2_SUB

  1348. 	KERNEL1x2_SUB

  1349. 

  1350. 	subs	counterL, counterL, #1

  1351. 	bgt	.Lsgemm_kernel_L2_M1_22

  1352. 

  1353. 

  1354. .Lsgemm_kernel_L2_M1_40:

  1355. 

  1356. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1357. 	ble	.Lsgemm_kernel_L2_M1_100

  1358. 

  1359. .Lsgemm_kernel_L2_M1_42:

  1360. 

  1361. 	KERNEL1x2_SUB

  1362. 

  1363. 	subs	counterL, counterL, #1

  1364. 	bgt	.Lsgemm_kernel_L2_M1_42

  1365. 

  1366. .Lsgemm_kernel_L2_M1_100:

  1367. 

  1368. 	SAVE1x2

  1369. 

  1370. 

  1371. .Lsgemm_kernel_L2_END:

  1372. 	add	origPB, origPB, origK, lsl #3	// B = B + K * 2 * 4

  1373. 

  1374. /******************************************************************************/

  1375. 

  1376. .Lsgemm_kernel_L1_BEGIN:

  1377. 

  1378. 	mov	counterJ , origN

  1379. 	tst	counterJ , #1

  1380. 	ble	.Lsgemm_kernel_L999 // done

  1381. 

  1382. 

  1383. 	mov	pCRow0, pC			// pCRow0 = C

  1384. 	add	pC , pC , LDC			// Update pC to point to next

  1385. 

  1386. 	mov	pA_0, origPA			// pA_0 = A

  1387. 

  1388. 

  1389. 

  1390. .Lsgemm_kernel_L1_M4_BEGIN:

  1391. 

  1392. 	mov	counterI, origM

  1393. 	asr 	counterI, counterI, #2		// counterI = counterI / 4

  1394. 	cmp	counterI, #0

  1395. 	ble	.Lsgemm_kernel_L1_M2_BEGIN

  1396. 

  1397. .Lsgemm_kernel_L1_M4_20:

  1398. 

  1399. 	INIT4x1

  1400. 

  1401. 	mov	pB, origPB

  1402. 	asr	counterL , origK, #3		// counterL = counterL / 8

  1403. 	cmp	counterL , #0

  1404. 	ble	.Lsgemm_kernel_L1_M4_40

  1405. 	.align 5

  1406. 

  1407. .Lsgemm_kernel_L1_M4_22:

  1408. 	KERNEL4x1_SUB

  1409. 	KERNEL4x1_SUB

  1410. 	KERNEL4x1_SUB

  1411. 	KERNEL4x1_SUB

  1412. 

  1413. 	KERNEL4x1_SUB

  1414. 	KERNEL4x1_SUB

  1415. 	KERNEL4x1_SUB

  1416. 	KERNEL4x1_SUB

  1417. 

  1418. 	subs	counterL, counterL, #1

  1419. 	bgt	.Lsgemm_kernel_L1_M4_22

  1420. 

  1421. 

  1422. .Lsgemm_kernel_L1_M4_40:

  1423. 

  1424. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1425. 	ble	.Lsgemm_kernel_L1_M4_100

  1426. 

  1427. .Lsgemm_kernel_L1_M4_42:

  1428. 

  1429. 	KERNEL4x1_SUB

  1430. 

  1431. 	subs	counterL, counterL, #1

  1432. 	bgt	.Lsgemm_kernel_L1_M4_42

  1433. 

  1434. .Lsgemm_kernel_L1_M4_100:

  1435. 

  1436. 	SAVE4x1

  1437. 

  1438. .Lsgemm_kernel_L1_M4_END:

  1439. 

  1440. 	subs	counterI, counterI, #1

  1441. 	bgt	.Lsgemm_kernel_L1_M4_20

  1442. 

  1443. 

  1444. .Lsgemm_kernel_L1_M2_BEGIN:

  1445. 

  1446. 	mov	counterI, origM

  1447. 	tst	counterI , #3

  1448. 	ble	.Lsgemm_kernel_L1_END

  1449. 

  1450. 	tst	counterI, #2			// counterI = counterI / 2

  1451. 	ble	.Lsgemm_kernel_L1_M1_BEGIN

  1452. 

  1453. .Lsgemm_kernel_L1_M2_20:

  1454. 

  1455. 	INIT2x1

  1456. 

  1457. 	mov	pB, origPB

  1458. 	asr 	counterL , origK, #3		// counterL = counterL / 8

  1459. 	cmp	counterL , #0

  1460. 	ble	.Lsgemm_kernel_L1_M2_40

  1461. 

  1462. .Lsgemm_kernel_L1_M2_22:

  1463. 

  1464. 	KERNEL2x1_SUB

  1465. 	KERNEL2x1_SUB

  1466. 	KERNEL2x1_SUB

  1467. 	KERNEL2x1_SUB

  1468. 

  1469. 	KERNEL2x1_SUB

  1470. 	KERNEL2x1_SUB

  1471. 	KERNEL2x1_SUB

  1472. 	KERNEL2x1_SUB

  1473. 

  1474. 	subs	counterL, counterL, #1

  1475. 	bgt	.Lsgemm_kernel_L1_M2_22

  1476. 

  1477. 

  1478. .Lsgemm_kernel_L1_M2_40:

  1479. 

  1480. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1481. 	ble	.Lsgemm_kernel_L1_M2_100

  1482. 

  1483. .Lsgemm_kernel_L1_M2_42:

  1484. 

  1485. 	KERNEL2x1_SUB

  1486. 

  1487. 	subs	counterL, counterL, #1

  1488. 	bgt	.Lsgemm_kernel_L1_M2_42

  1489. 

  1490. .Lsgemm_kernel_L1_M2_100:

  1491. 

  1492. 	SAVE2x1

  1493. 

  1494. .Lsgemm_kernel_L1_M2_END:

  1495. 

  1496. 

  1497. .Lsgemm_kernel_L1_M1_BEGIN:

  1498. 

  1499. 	tst	counterI, #1			// counterI = counterI % 2

  1500. 	ble	.Lsgemm_kernel_L1_END

  1501. 

  1502. .Lsgemm_kernel_L1_M1_20:

  1503. 

  1504. 	INIT1x1

  1505. 

  1506. 	mov	pB, origPB

  1507. 	asr 	counterL , origK, #3		// counterL = counterL / 8

  1508. 	cmp	counterL , #0

  1509. 	ble	.Lsgemm_kernel_L1_M1_40

  1510. 

  1511. .Lsgemm_kernel_L1_M1_22:

  1512. 	KERNEL1x1_SUB

  1513. 	KERNEL1x1_SUB

  1514. 	KERNEL1x1_SUB

  1515. 	KERNEL1x1_SUB

  1516. 

  1517. 	KERNEL1x1_SUB

  1518. 	KERNEL1x1_SUB

  1519. 	KERNEL1x1_SUB

  1520. 	KERNEL1x1_SUB

  1521. 

  1522. 	subs	counterL, counterL, #1

  1523. 	bgt	.Lsgemm_kernel_L1_M1_22

  1524. 

  1525. 

  1526. .Lsgemm_kernel_L1_M1_40:

  1527. 

  1528. 	ands	counterL , origK, #7		// counterL = counterL % 8

  1529. 	ble	.Lsgemm_kernel_L1_M1_100

  1530. 

  1531. .Lsgemm_kernel_L1_M1_42:

  1532. 

  1533. 	KERNEL1x1_SUB

  1534. 

  1535. 	subs	counterL, counterL, #1

  1536. 	bgt	.Lsgemm_kernel_L1_M1_42

  1537. 

  1538. .Lsgemm_kernel_L1_M1_100:

  1539. 

  1540. 	SAVE1x1

  1541. 

  1542. 

  1543. .Lsgemm_kernel_L1_END:

  1544. 

  1545. 

  1546. .Lsgemm_kernel_L999:

  1547. 	mov	x0, #0				// set return value

  1548. 	ldp	d8, d9, [sp, #(0 * 16)]

  1549. 	ldp	d10, d11, [sp, #(1 * 16)]

  1550. 	ldp	d12, d13, [sp, #(2 * 16)]

  1551. 	ldp	d14, d15, [sp, #(3 * 16)]

  1552. 	ldp	d16, d17, [sp, #(4 * 16)]

  1553. 	ldp	x18, x19, [sp, #(5 * 16)]

  1554. 	ldp	x20, x21, [sp, #(6 * 16)]

  1555. 	ldp	x22, x23, [sp, #(7 * 16)]

  1556. 	ldp	x24, x25, [sp, #(8 * 16)]

  1557. 	ldp	x26, x27, [sp, #(9 * 16)]

  1558. 	ldr	x28, [sp, #(10 * 16)]

  1559. 	add	sp, sp, #(11*16)

  1560. 	ret

  1561. 

  1562. 	EPILOGUE

  1563.

OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.