commit ac654f1fd44e19e7fffbd6e364233b2080120b05
author Frank Barchard <fbarchard@google.com> Mon Jan 24 23:51:04 2022 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Mon Jan 24 23:52:04 2022 -0800
tree 6e2d319985779916f644700dea920304bb63fd1f
parent 364598a012f6a3190a5a34bac41733a8ef8495ec

QC8 4x8 dot product GEMM AArch32 microkernel for Cortex A55

- FP32 quantization
- reduce stack push by 16 bytes improving QS8 as well as QC8 GEMM.

PiperOrigin-RevId: 424004581

BUILD.bazel

diff --git a/BUILD.bazel b/BUILD.bazel
index 2e7e083..9238d23 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -7052,6 +7052,7 @@
     "src/qc8-gemm/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-cortex-a55.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",