Merge pull request #717 from shi510:fix_build_err
PiperOrigin-RevId: 316357949
diff --git a/CMakeLists.txt b/CMakeLists.txt
old mode 100644
new mode 100755
index 5ae121a..55927c0
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -184,7 +184,7 @@
src/subgraph/argmax-pooling-2d.c
src/subgraph/average-pooling-2d.c
src/subgraph/bankers-rounding.c
- src/subgraph/ceil.c
+ src/subgraph/ceiling.c
src/subgraph/clamp.c
src/subgraph/convolution-2d.c
src/subgraph/deconvolution-2d.c
@@ -307,9 +307,9 @@
src/f32-igemm/gen/2x4-minmax-scalar.c
src/f32-igemm/gen/4x2-minmax-scalar.c
src/f32-igemm/gen/4x4-minmax-scalar.c
- src/f32-lrelu/gen/scalar-x1.c
- src/f32-lrelu/gen/scalar-x2.c
- src/f32-lrelu/gen/scalar-x4.c
+ src/f32-vlrelu/gen/scalar-x1.c
+ src/f32-vlrelu/gen/scalar-x2.c
+ src/f32-vlrelu/gen/scalar-x4.c
src/f32-maxpool/9p8x-minmax-scalar-c1.c
src/f32-pavgpool/9p8x-minmax-scalar-c1.c
src/f32-pavgpool/9x-minmax-scalar-c1.c
@@ -539,8 +539,8 @@
src/f32-igemm/gen/6x8-minmax-psimd-loadsplat.c
src/f32-igemm/gen/6x8-minmax-psimd-splat.c
src/f32-igemm/gen/6x8s4-minmax-psimd.c
- src/f32-lrelu/gen/psimd-x4.c
- src/f32-lrelu/gen/psimd-x8.c
+ src/f32-vlrelu/gen/psimd-x4.c
+ src/f32-vlrelu/gen/psimd-x8.c
src/f32-maxpool/9p8x-minmax-psimd-c4.c
src/f32-pavgpool/9p8x-minmax-psimd-c4.c
src/f32-pavgpool/9x-minmax-psimd-c4.c
@@ -716,8 +716,8 @@
src/f32-igemm/gen/4x8s4-minmax-neon.c
src/f32-igemm/gen/6x8s4-minmax-neon.c
src/f32-igemm/gen/8x8s4-minmax-neon.c
- src/f32-lrelu/gen/neon-x4.c
- src/f32-lrelu/gen/neon-x8.c
+ src/f32-vlrelu/gen/neon-x4.c
+ src/f32-vlrelu/gen/neon-x8.c
src/f32-maxpool/9p8x-minmax-neon-c4.c
src/f32-pavgpool/9p8x-minmax-neon-c4.c
src/f32-pavgpool/9x-minmax-neon-c4.c
@@ -1236,8 +1236,8 @@
src/f32-igemm/gen/4x8-minmax-sse-dup.c
src/f32-igemm/gen/4x8-minmax-sse-load1.c
src/f32-igemm/gen/4x8s4-minmax-sse.c
- src/f32-lrelu/gen/sse-x4.c
- src/f32-lrelu/gen/sse-x8.c
+ src/f32-vlrelu/gen/sse-x4.c
+ src/f32-vlrelu/gen/sse-x8.c
src/f32-maxpool/9p8x-minmax-sse-c4.c
src/f32-pavgpool/9p8x-minmax-sse-c4.c
src/f32-pavgpool/9x-minmax-sse-c4.c
@@ -1435,8 +1435,8 @@
src/f32-igemm/gen/3x16-minmax-avx-broadcast.c
src/f32-igemm/gen/4x16-minmax-avx-broadcast.c
src/f32-igemm/gen/5x16-minmax-avx-broadcast.c
- src/f32-lrelu/gen/avx-x8.c
- src/f32-lrelu/gen/avx-x16.c
+ src/f32-vlrelu/gen/avx-x8.c
+ src/f32-vlrelu/gen/avx-x16.c
src/f32-prelu/gen/avx-2x8.c
src/f32-prelu/gen/avx-2x16.c
src/f32-rmax/avx.c
@@ -1686,8 +1686,8 @@
src/f32-igemm/gen/6x16-minmax-avx512f-broadcast.c
src/f32-igemm/gen/7x16-minmax-avx512f-broadcast.c
src/f32-igemm/gen/8x16-minmax-avx512f-broadcast.c
- src/f32-lrelu/gen/avx512f-x16.c
- src/f32-lrelu/gen/avx512f-x32.c
+ src/f32-vlrelu/gen/avx512f-x16.c
+ src/f32-vlrelu/gen/avx512f-x32.c
src/f32-prelu/gen/avx512f-2x16.c
src/f32-prelu/gen/avx512f-2x32.c
src/f32-raddexpminusmax/gen/avx512f-p5-scalef-x128.c
@@ -2312,7 +2312,7 @@
CXX_STANDARD_REQUIRED YES
CXX_EXTENSIONS NO)
TARGET_INCLUDE_DIRECTORIES(global-average-pooling-nwc-test PRIVATE src test)
- TARGET_LINK_LIBRARIES(global-average-pooling-nwc-test PRIVATE XNNPACK gtest gtest_main)
+ TARGET_LINK_LIBRARIES(global-average-pooling-nwc-test PRIVATE XNNPACK fp16 gtest gtest_main)
ADD_TEST(global-average-pooling-nwc-test global-average-pooling-nwc-test)
ADD_EXECUTABLE(global-average-pooling-ncw-test test/global-average-pooling-ncw.cc)
@@ -2321,7 +2321,7 @@
CXX_STANDARD_REQUIRED YES
CXX_EXTENSIONS NO)
TARGET_INCLUDE_DIRECTORIES(global-average-pooling-ncw-test PRIVATE src test)
- TARGET_LINK_LIBRARIES(global-average-pooling-ncw-test PRIVATE XNNPACK gtest gtest_main)
+ TARGET_LINK_LIBRARIES(global-average-pooling-ncw-test PRIVATE XNNPACK fp16 gtest gtest_main)
ADD_TEST(global-average-pooling-ncw-test global-average-pooling-ncw-test)
ADD_EXECUTABLE(hardswish-nc-test test/hardswish-nc.cc)
diff --git a/src/f32-dwconv-chw/5x5p2-psimd.c b/src/f32-dwconv-chw/5x5p2-psimd.c
old mode 100644
new mode 100755
index f2fa401..03a5ab1
--- a/src/f32-dwconv-chw/5x5p2-psimd.c
+++ b/src/f32-dwconv-chw/5x5p2-psimd.c
@@ -73,7 +73,7 @@
const size_t padded_input_height = input_height + padding_top + 2 /* padding_bottom */;
const size_t output_height = padded_input_height - 5 + 1;
- const psimd_s32 vmask = psimd_load_f32(params->scalar.mask);
+ const psimd_s32 vmask = psimd_load_s32(params->scalar.mask);
const psimd_f32 vmax = psimd_load_splat_f32(¶ms->scalar.max);
const psimd_f32 vmin = psimd_load_splat_f32(¶ms->scalar.min);