commit 18630de490a8b6840080e9bd621e556be096a51c
author Marat Dukhan <maratek@google.com> Wed Jun 02 22:20:01 2021 -0700
committer XNNPACK Team <xnnpack-github-robot@google.com> Wed Jun 02 22:20:47 2021 -0700
tree e498f8db4bd26c18b712a7e9586d93a0b6d90355
parent 801d2c2d7e3576b368d2f841b357adebc8ac1527

QS8 NEONDOT GEMM/IGEMM microkernels with FP32 requantization

PiperOrigin-RevId: 377216784

scripts/generate-qs8-igemm.sh

diff --git a/scripts/generate-qs8-igemm.sh b/scripts/generate-qs8-igemm.sh
index 76f7e64..a7df436 100755
--- a/scripts/generate-qs8-igemm.sh
+++ b/scripts/generate-qs8-igemm.sh
@@ -107,14 +107,23 @@
 tools/xngen src/qs8-igemm/c16-neon-mlal-padal.c.in -D MR=4 -D NR=16 -o src/qs8-igemm/gen/4x16c16-minmax-gemmlowp-neon-mlal-padal.c
 
 ### C4 micro-kernels
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=8  -o src/qs8-igemm/gen/1x8c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=8  -o src/qs8-igemm/gen/4x8c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=8  -o src/qs8-igemm/gen/6x8c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=8  -o src/qs8-igemm/gen/8x8c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=16 -o src/qs8-igemm/gen/1x16c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=16 -o src/qs8-igemm/gen/4x16c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=16 -o src/qs8-igemm/gen/6x16c4-minmax-gemmlowp-neondot.c
-tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=16 -o src/qs8-igemm/gen/8x16c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=8  -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/1x8c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=8  -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/4x8c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=8  -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/6x8c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=8  -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/8x8c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=16 -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/1x16c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=16 -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/4x16c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=16 -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/6x16c4-minmax-gemmlowp-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=16 -D REQUANTIZATION=GEMMLOWP -o src/qs8-igemm/gen/8x16c4-minmax-gemmlowp-neondot.c
+
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=8  -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/1x8c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=8  -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/4x8c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=8  -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/6x8c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=8  -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/8x8c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=1  -D NR=16 -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/1x16c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=4  -D NR=16 -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/4x16c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=6  -D NR=16 -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/6x16c4-minmax-fp32-neondot.c
+tools/xngen src/qs8-igemm/MRxNRc4-neondot.c.in -D MR=8  -D NR=16 -D REQUANTIZATION=FP32     -o src/qs8-igemm/gen/8x16c4-minmax-fp32-neondot.c
 
 
 ############################### AArch64 assembly ##############################