commit 71855ee259c9215e3f9deeaff82a7edb270b3357
author Marat Dukhan <maratek@google.com> Tue May 25 19:05:06 2021 -0700
committer XNNPACK Team <xnnpack-github-robot@google.com> Tue May 25 19:06:19 2021 -0700
tree f51bef17f46c2905edda74c0b59f3413c3a37811
parent d4c7d8218ba910561375a71f1dea2d7c76bc6122

Support FP32 requantization in AVX512 QS8 microkernels

PiperOrigin-RevId: 375845159

CMakeLists.txt

diff --git a/CMakeLists.txt b/CMakeLists.txt
index f9840bb..574dc7e 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2900,17 +2900,29 @@
 
 SET(XNNPACK_AVX512SKX_MICROKERNEL_SRCS
   src/qs8-dwconv/gen/up16x9-minmax-gemmlowp-avx512skx-mul32.c
+  src/qs8-dwconv/gen/up16x9-minmax-fp32-avx512skx-mul32.c
   src/qs8-dwconv/gen/up16x25-minmax-gemmlowp-avx512skx-mul32.c
+  src/qs8-dwconv/gen/up16x25-minmax-fp32-avx512skx-mul32.c
   src/qs8-dwconv/gen/up32x9-minmax-gemmlowp-avx512skx-mul32.c
+  src/qs8-dwconv/gen/up32x9-minmax-fp32-avx512skx-mul32.c
   src/qs8-dwconv/gen/up32x25-minmax-gemmlowp-avx512skx-mul32.c
+  src/qs8-dwconv/gen/up32x25-minmax-fp32-avx512skx-mul32.c
   src/qs8-gemm/gen/1x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-gemm/gen/1x16c8-minmax-fp32-avx512skx.c
   src/qs8-gemm/gen/2x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-gemm/gen/2x16c8-minmax-fp32-avx512skx.c
   src/qs8-gemm/gen/3x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-gemm/gen/3x16c8-minmax-fp32-avx512skx.c
   src/qs8-gemm/gen/4x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-gemm/gen/4x16c8-minmax-fp32-avx512skx.c
   src/qs8-igemm/gen/1x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-igemm/gen/1x16c8-minmax-fp32-avx512skx.c
   src/qs8-igemm/gen/2x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-igemm/gen/2x16c8-minmax-fp32-avx512skx.c
   src/qs8-igemm/gen/3x16c8-minmax-gemmlowp-avx512skx.c
-  src/qs8-igemm/gen/4x16c8-minmax-gemmlowp-avx512skx.c)
+  src/qs8-igemm/gen/3x16c8-minmax-fp32-avx512skx.c
+  src/qs8-igemm/gen/4x16c8-minmax-gemmlowp-avx512skx.c
+  src/qs8-igemm/gen/4x16c8-minmax-fp32-avx512skx.c)
 
 SET(XNNPACK_AARCH32_ASM_MICROKERNEL_SRCS
   src/f32-gemm/4x4-aarch32-vfp-ld64.S