commit c5aa242a08b18dca7e14f665cea0a6a1680dd8e4
author Marat Dukhan <maratek@google.com> Wed Dec 01 00:15:19 2021 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> Wed Dec 01 00:16:11 2021 -0800
tree fdcd6650c8cb38eade26a1ebc2cf024718641c3f
parent 0d6a1194ff36f59c79089336d7a3b446c3c6a39d

F32->QS8 and F32->QU8 microkernels for SSE

PiperOrigin-RevId: 413337694

CMakeLists.txt

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 73cf57e..d538964 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3278,6 +3278,14 @@
   src/f32-igemm/gen/5x8-minmax-sse2-dup.c
   src/f32-prelu/gen/sse2-2x4.c
   src/f32-prelu/gen/sse2-2x8.c
+  src/f32-qs8-vcvt/gen/vcvt-sse2-x8.c
+  src/f32-qs8-vcvt/gen/vcvt-sse2-x16.c
+  src/f32-qs8-vcvt/gen/vcvt-sse2-x24.c
+  src/f32-qs8-vcvt/gen/vcvt-sse2-x32.c
+  src/f32-qu8-vcvt/gen/vcvt-sse2-x8.c
+  src/f32-qu8-vcvt/gen/vcvt-sse2-x16.c
+  src/f32-qu8-vcvt/gen/vcvt-sse2-x24.c
+  src/f32-qu8-vcvt/gen/vcvt-sse2-x32.c
   src/f32-raddstoreexpminusmax/gen/sse2-p5-x4.c
   src/f32-raddstoreexpminusmax/gen/sse2-p5-x8-acc2.c
   src/f32-raddstoreexpminusmax/gen/sse2-p5-x8.c
@@ -3644,6 +3652,10 @@
   src/f32-f16-vcvt/gen/vcvt-sse41-x32.c
   src/f32-prelu/gen/sse41-2x4.c
   src/f32-prelu/gen/sse41-2x8.c
+  src/f32-qs8-vcvt/gen/vcvt-sse41-x8.c
+  src/f32-qs8-vcvt/gen/vcvt-sse41-x16.c
+  src/f32-qs8-vcvt/gen/vcvt-sse41-x24.c
+  src/f32-qs8-vcvt/gen/vcvt-sse41-x32.c
   src/f32-velu/gen/velu-sse41-rr2-lut16-p3-x4.c
   src/f32-velu/gen/velu-sse41-rr2-lut16-p3-x8.c
   src/f32-velu/gen/velu-sse41-rr2-lut16-p3-x12.c
@@ -6510,6 +6522,24 @@
   TARGET_LINK_LIBRARIES(f32-f16-vcvt-test PRIVATE cpuinfo fp16 pthreadpool gtest gtest_main)
   ADD_TEST(f32-f16-vcvt-test f32-f16-vcvt-test)
 
+  ADD_EXECUTABLE(f32-qs8-vcvt-test test/f32-f16-vcvt.cc $<TARGET_OBJECTS:all_microkernels>)
+  SET_TARGET_PROPERTIES(f32-qs8-vcvt-test PROPERTIES
+    CXX_STANDARD 11
+    CXX_STANDARD_REQUIRED YES
+    CXX_EXTENSIONS YES)
+  TARGET_INCLUDE_DIRECTORIES(f32-qs8-vcvt-test PRIVATE include src test)
+  TARGET_LINK_LIBRARIES(f32-qs8-vcvt-test PRIVATE cpuinfo fp16 pthreadpool gtest gtest_main)
+  ADD_TEST(f32-qs8-vcvt-test f32-qs8-vcvt-test)
+
+  ADD_EXECUTABLE(f32-qu8-vcvt-test test/f32-f16-vcvt.cc $<TARGET_OBJECTS:all_microkernels>)
+  SET_TARGET_PROPERTIES(f32-qu8-vcvt-test PROPERTIES
+    CXX_STANDARD 11
+    CXX_STANDARD_REQUIRED YES
+    CXX_EXTENSIONS YES)
+  TARGET_INCLUDE_DIRECTORIES(f32-qu8-vcvt-test PRIVATE include src test)
+  TARGET_LINK_LIBRARIES(f32-qu8-vcvt-test PRIVATE cpuinfo fp16 pthreadpool gtest gtest_main)
+  ADD_TEST(f32-qu8-vcvt-test f32-qu8-vcvt-test)
+
   ADD_EXECUTABLE(f32-gavgpool-cw-test test/f32-gavgpool-cw.cc $<TARGET_OBJECTS:all_microkernels>)
   SET_TARGET_PROPERTIES(f32-gavgpool-cw-test PROPERTIES
     CXX_STANDARD 11