Enable QU8 3x8 lane for AArch32
- Was 2x16, now 3x8. Less register spill improves performance.
PiperOrigin-RevId: 421214812
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7af9d07..e02cfa8 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1076,23 +1076,23 @@
src/qs8-f32-vcvt/gen/vcvt-scalar-x2.c
src/qs8-f32-vcvt/gen/vcvt-scalar-x3.c
src/qs8-f32-vcvt/gen/vcvt-scalar-x4.c
- src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c1.c
src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-fmagic-c1.c
- src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-lrintf-c1.c
- src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c2.c
src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-fmagic-c2.c
- src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-lrintf-c2.c
- src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c4.c
src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-fmagic-c4.c
+ src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c1.c
+ src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c2.c
+ src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-imagic-c4.c
+ src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-lrintf-c1.c
+ src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-lrintf-c2.c
src/qs8-gavgpool/gen/7p7x-minmax-fp32-scalar-lrintf-c4.c
- src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c1.c
src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-fmagic-c1.c
- src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-lrintf-c1.c
- src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c2.c
src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-fmagic-c2.c
- src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-lrintf-c2.c
- src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c4.c
src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-fmagic-c4.c
+ src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c1.c
+ src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c2.c
+ src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-imagic-c4.c
+ src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-lrintf-c1.c
+ src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-lrintf-c2.c
src/qs8-gavgpool/gen/7x-minmax-fp32-scalar-lrintf-c4.c
src/qs8-gemm/gen/1x2-minmax-fp32-scalar-fmagic.c
src/qs8-gemm/gen/1x2-minmax-fp32-scalar-imagic.c
@@ -1400,11 +1400,13 @@
src/qu8-f32-vcvt/gen/vcvt-neon-x32.c
src/qu8-gavgpool/7p7x-minmax-neon-c8.c
src/qu8-gavgpool/7x-minmax-neon-c8.c
+ src/qu8-gemm/gen/1x8-minmax-rndnu-neon-mlal-lane.c
src/qu8-gemm/gen/1x16-minmax-rndnu-neon-mlal-lane.c
- src/qu8-gemm/gen/2x16-minmax-rndnu-neon-mlal-lane.c
+ src/qu8-gemm/gen/3x8-minmax-rndnu-neon-mlal-lane.c
src/qu8-gemm/gen/4x16-minmax-rndnu-neon-mlal-lane.c
+ src/qu8-igemm/gen/1x8-minmax-rndnu-neon-mlal-lane.c
src/qu8-igemm/gen/1x16-minmax-rndnu-neon-mlal-lane.c
- src/qu8-igemm/gen/2x16-minmax-rndnu-neon-mlal-lane.c
+ src/qu8-igemm/gen/3x8-minmax-rndnu-neon-mlal-lane.c
src/qu8-igemm/gen/4x16-minmax-rndnu-neon-mlal-lane.c
src/qu8-vadd/gen/minmax-neon-ld64-x16.c
src/qu8-vadd/gen/minmax-neon-ld64-x32.c