aboutsummaryrefslogtreecommitdiff
path: root/scripts
diff options
context:
space:
mode:
authorFrank Barchard <fbarchard@google.com>2021-01-22 14:08:02 -0800
committerXNNPACK Team <xnnpack-github-robot@google.com>2021-01-22 14:09:03 -0800
commit2302ffd7d9b2a68ae622dffb2aaa73dc57d49915 (patch)
tree1302d362131207b636ddea56f4aa0a6df86712df /scripts
parentcadd422eb2642e6d1e9c73bead5ca3cde8c4e011 (diff)
downloadXNNPACK-2302ffd7d9b2a68ae622dffb2aaa73dc57d49915.tar.gz
QS8 Neon GEMM microkernel with 8 bit multiply and vpadal to accumulate
PiperOrigin-RevId: 353315852
Diffstat (limited to 'scripts')
-rwxr-xr-xscripts/generate-qs8-gemm.sh10
1 files changed, 10 insertions, 0 deletions
diff --git a/scripts/generate-qs8-gemm.sh b/scripts/generate-qs8-gemm.sh
index f6cf0bec8..4cf3fc5fb 100755
--- a/scripts/generate-qs8-gemm.sh
+++ b/scripts/generate-qs8-gemm.sh
@@ -44,6 +44,16 @@ tools/xngen src/qs8-gemm/neon-mull-addw-dup.c.in -D MR=2 -D NR=16 -o src/qs8-gem
tools/xngen src/qs8-gemm/neon-mull-addw-dup.c.in -D MR=3 -D NR=16 -o src/qs8-gemm/gen/3x16-minmax-neon-mull-addw-dup.c
tools/xngen src/qs8-gemm/neon-mull-addw-dup.c.in -D MR=4 -D NR=16 -o src/qs8-gemm/gen/4x16-minmax-neon-mull-addw-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=1 -D NR=8 -o src/qs8-gemm/gen/1x8c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=2 -D NR=8 -o src/qs8-gemm/gen/2x8c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=3 -D NR=8 -o src/qs8-gemm/gen/3x8c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=4 -D NR=8 -o src/qs8-gemm/gen/4x8c2-minmax-neon-mull-padal-dup.c
+
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=1 -D NR=16 -o src/qs8-gemm/gen/1x16c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=2 -D NR=16 -o src/qs8-gemm/gen/2x16c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=3 -D NR=16 -o src/qs8-gemm/gen/3x16c2-minmax-neon-mull-padal-dup.c
+tools/xngen src/qs8-gemm/c2-neon-mull-padal-dup.c.in -D MR=4 -D NR=16 -o src/qs8-gemm/gen/4x16c2-minmax-neon-mull-padal-dup.c
+
### C4 micro-kernels
tools/xngen src/qs8-gemm/MRxNRc4-neondot.c.in -D MR=1 -D NR=8 -o src/qs8-gemm/gen/1x8c4-minmax-neondot.c
tools/xngen src/qs8-gemm/MRxNRc4-neondot.c.in -D MR=4 -D NR=8 -o src/qs8-gemm/gen/4x8c4-minmax-neondot.c