Enable QC8 AArch32 4x8 lane GEMM/IGEMM assembly microkernels for ARMv7 NEON
- FP32 Quantization for QC8 uses ARMv7 compatible code.
- Replace C2S4 intrinsics microkernel with 4x8 lane assembly.
PiperOrigin-RevId: 421946062
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 71767a6..7dd0272 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -377,8 +377,10 @@
src/qc8-dwconv/gen/up2x9-minmax-fp32-scalar-fmagic.c
src/qc8-dwconv/gen/up2x25-minmax-fp32-scalar-fmagic.c
src/qc8-gemm/gen/1x2-minmax-fp32-scalar-fmagic.c
+ src/qc8-gemm/gen/1x8-minmax-fp32-neon-mlal-lane.c
src/qc8-gemm/gen/2x2-minmax-fp32-scalar-fmagic.c
src/qc8-igemm/gen/1x2-minmax-fp32-scalar-fmagic.c
+ src/qc8-igemm/gen/1x8-minmax-fp32-neon-mlal-lane.c
src/qc8-igemm/gen/2x2-minmax-fp32-scalar-fmagic.c
src/qs8-dwconv/gen/up1x9-minmax-fp32-scalar-fmagic.c
src/qs8-dwconv/gen/up1x25-minmax-fp32-scalar-fmagic.c
@@ -5756,11 +5758,11 @@
src/f32-igemm/gen/4x8-minmax-aarch32-neon-prfm-cortex-a75.S
src/qc8-gemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-ld64.S
src/qc8-gemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-prfm-ld64.S
- src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-ld64.S
- src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-prfm-ld64.S
src/qc8-gemm/gen/4x8-minmax-fp32-aarch32-neonv8-mlal-lane-ld64.S
src/qc8-gemm/gen/4x8-minmax-fp32-aarch32-neonv8-mlal-lane-prfm-ld64.S
src/qc8-gemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S
+ src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-ld64.S
+ src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neon-mlal-lane-prfm-ld64.S
src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neonv8-mlal-lane-ld64.S
src/qc8-igemm/gen/4x8-minmax-fp32-aarch32-neonv8-mlal-lane-prfm-ld64.S
src/qc8-igemm/gen/4x8c4-minmax-fp32-aarch32-neondot-ld64.S