Refactor microkernel lists in BUILD and CMakeLists.txt
Unify naming of microkernel lists across the two build systems
PiperOrigin-RevId: 387120593
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4127bf5..3a2295a 100755
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -160,7 +160,7 @@
ENDIF()
# ---[ XNNPACK library
-SET(XNNPACK_OPERATOR_SRCS
+SET(OPERATOR_SRCS
src/operators/argmax-pooling-nhwc.c
src/operators/average-pooling-nhwc.c
src/operators/binary-elementwise-nd.c
@@ -184,7 +184,7 @@
src/operators/unary-elementwise-nc.c
src/operators/unpooling-nhwc.c)
-SET(XNNPACK_SUBGRAPH_SRCS
+SET(SUBGRAPH_SRCS
src/subgraph/abs.c
src/subgraph/add2.c
src/subgraph/argmax-pooling-2d.c
@@ -220,13 +220,13 @@
src/subgraph/subtract.c
src/subgraph/unpooling-2d.c)
-SET(XNNPACK_LOGGING_SRCS
+SET(LOGGING_SRCS
src/datatype-strings.c
src/operator-strings.c
src/subgraph-strings.c)
-SET(XNNPACK_COLD_SRCS ${XNNPACK_OPERATOR_SRCS} ${XNNPACK_SUBGRAPH_SRCS} ${XNNPACK_LOGGING_SRCS})
-LIST(APPEND XNNPACK_COLD_SRCS
+SET(COLD_SRCS ${OPERATOR_SRCS} ${SUBGRAPH_SRCS} ${LOGGING_SRCS})
+LIST(APPEND COLD_SRCS
src/allocator.c
src/init.c
src/memory-planner.c
@@ -235,12 +235,12 @@
src/subgraph.c
src/tensor.c)
-SET(XNNPACK_HOT_SRCS
+SET(HOT_SRCS
src/indirection.c
src/operator-run.c
src/packing.c)
-SET(XNNPACK_TABLE_SRCS
+SET(TABLE_SRCS
src/tables/exp2-k-over-64.c
src/tables/exp2-k-over-2048.c
src/tables/exp2minus-k-over-4.c
@@ -249,7 +249,7 @@
src/tables/exp2minus-k-over-64.c
src/tables/exp2minus-k-over-2048.c)
-SET(XNNPACK_SCALAR_MICROKERNEL_SRCS
+SET(SCALAR_MICROKERNEL_SRCS
src/f32-argmaxpool/4x-scalar-c1.c
src/f32-argmaxpool/9p8x-scalar-c1.c
src/f32-argmaxpool/9x-scalar-c1.c
@@ -848,7 +848,7 @@
src/x32-zip/xm-scalar.c
src/xx-copy/memcpy.c)
-SET(XNNPACK_NEON_MICROKERNEL_SRCS
+SET(NEON_MICROKERNEL_SRCS
src/f32-argmaxpool/4x-neon-c4.c
src/f32-argmaxpool/9p8x-neon-c4.c
src/f32-argmaxpool/9x-neon-c4.c
@@ -1434,7 +1434,7 @@
src/x32-zip/x4-neon.c
src/x32-zip/xm-neon.c)
-SET(XNNPACK_NEONFMA_MICROKERNEL_SRCS
+SET(NEONFMA_MICROKERNEL_SRCS
src/f32-dwconv/gen/up4x4-minmax-neonfma-acc2.c
src/f32-dwconv/gen/up4x4-minmax-neonfma.c
src/f32-dwconv/gen/up4x9-minmax-neonfma-acc2.c
@@ -1636,7 +1636,7 @@
src/math/sqrt-neonfma-nr2fma1adj.c
src/math/sqrt-neonfma-nr3fma.c)
-SET(XNNPACK_NEONV8_MICROKERNEL_SRCS
+SET(NEONV8_MICROKERNEL_SRCS
src/f32-vrnd/gen/vrndd-neonv8-x4.c
src/f32-vrnd/gen/vrndd-neonv8-x8.c
src/f32-vrnd/gen/vrndne-neonv8-x4.c
@@ -1714,7 +1714,7 @@
src/qu8-igemm/gen/1x16-minmax-fp32-neonv8-mlal-lane.c
src/qu8-igemm/gen/4x16-minmax-fp32-neonv8-mlal-lane.c)
-SET(XNNPACK_AARCH64_NEONFMA_MICROKERNEL_SRCS
+SET(AARCH64_NEONFMA_MICROKERNEL_SRCS
src/f32-conv-hwc/gen/3x3s2p0p1c3x4-neonfma-2x1.c
src/f32-conv-hwc/gen/3x3s2p0p1c3x4-neonfma-2x2.c
src/f32-conv-hwc/gen/3x3s2p0p1c3x8-neonfma-2x1.c
@@ -1828,7 +1828,7 @@
src/math/sigmoid-neonfma-rr2-lut2048-p1-div.c
src/math/sigmoid-neonfma-rr2-p5-div.c)
-SET(XNNPACK_AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS
+SET(AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS
src/f16-dwconv/gen/up8x4-minmax-neonfp16arith-acc2.c
src/f16-dwconv/gen/up8x4-minmax-neonfp16arith.c
src/f16-dwconv/gen/up8x9-minmax-neonfp16arith-acc2.c
@@ -1914,7 +1914,7 @@
src/f16-vrelu/gen/vrelu-neonfp16arith-x8.c
src/f16-vrelu/gen/vrelu-neonfp16arith-x16.c)
-SET(XNNPACK_NEONDOT_MICROKERNEL_SRCS
+SET(NEONDOT_MICROKERNEL_SRCS
src/qc8-gemm/gen/1x8c4-minmax-fp32-neondot.c
src/qc8-gemm/gen/1x16c4-minmax-fp32-neondot.c
src/qc8-gemm/gen/4x8c4-minmax-fp32-neondot.c
@@ -1972,7 +1972,7 @@
src/qs8-igemm/gen/8x16c4-minmax-fp32-neondot.c
src/qs8-igemm/gen/8x16c4-minmax-gemmlowp-neondot.c)
-SET(XNNPACK_SSE_MICROKERNEL_SRCS
+SET(SSE_MICROKERNEL_SRCS
src/f32-avgpool/9p8x-minmax-sse-c4.c
src/f32-avgpool/9x-minmax-sse-c4.c
src/f32-conv-hwc2chw/3x3s2p1c3x4-sse-1x1.c
@@ -2147,7 +2147,7 @@
src/x32-packx/x4-sse.c
src/x32-pad/sse.c)
-SET(XNNPACK_SSE2_MICROKERNEL_SRCS
+SET(SSE2_MICROKERNEL_SRCS
src/f32-argmaxpool/4x-sse2-c4.c
src/f32-argmaxpool/9p8x-sse2-c4.c
src/f32-argmaxpool/9x-sse2-c4.c
@@ -2396,7 +2396,7 @@
src/x32-zip/x4-sse2.c
src/x32-zip/xm-sse2.c)
-SET(XNNPACK_SSSE3_MICROKERNEL_SRCS
+SET(SSSE3_MICROKERNEL_SRCS
src/f32-dwconv2d-chw/gen/3x3p1-minmax-ssse3-1x4-acc2.c
src/f32-dwconv2d-chw/gen/3x3p1-minmax-ssse3-1x4-acc3.c
src/f32-dwconv2d-chw/gen/3x3p1-minmax-ssse3-1x4-acc4.c
@@ -2451,7 +2451,7 @@
src/qu8-requantization/gemmlowp-ssse3.c
src/qu8-requantization/rndna-ssse3.c)
-SET(XNNPACK_SSE41_MICROKERNEL_SRCS
+SET(SSE41_MICROKERNEL_SRCS
src/f32-prelu/gen/sse41-2x4.c
src/f32-prelu/gen/sse41-2x8.c
src/f32-velu/gen/velu-sse41-rr2-lut16-p3-x4.c
@@ -2681,7 +2681,7 @@
src/qu8-vaddc/gen/minmax-sse41-mul32-ld32-x8.c
src/qu8-vaddc/gen/minmax-sse41-mul32-ld32-x16.c)
-SET(XNNPACK_AVX_MICROKERNEL_SRCS
+SET(AVX_MICROKERNEL_SRCS
src/f32-dwconv/gen/up8x4-minmax-avx-acc2.c
src/f32-dwconv/gen/up8x4-minmax-avx.c
src/f32-dwconv/gen/up8x9-minmax-avx-acc2.c
@@ -2993,7 +2993,7 @@
src/qu8-vaddc/gen/minmax-avx-mul32-ld32-x8.c
src/qu8-vaddc/gen/minmax-avx-mul32-ld32-x16.c)
-SET(XNNPACK_XOP_MICROKERNEL_SRCS
+SET(XOP_MICROKERNEL_SRCS
src/qc8-dwconv/gen/up8x9-minmax-fp32-xop-mul16-add16.c
src/qc8-dwconv/gen/up8x9-minmax-fp32-xop-mul32.c
src/qc8-dwconv/gen/up8x25-minmax-fp32-xop-mul16-add16.c
@@ -3128,7 +3128,7 @@
src/qu8-vaddc/gen/minmax-xop-mul32-ld32-x8.c
src/qu8-vaddc/gen/minmax-xop-mul32-ld32-x16.c)
-SET(XNNPACK_FMA3_MICROKERNEL_SRCS
+SET(FMA3_MICROKERNEL_SRCS
src/f32-dwconv/gen/up8x4-minmax-fma3-acc2.c
src/f32-dwconv/gen/up8x4-minmax-fma3.c
src/f32-dwconv/gen/up8x9-minmax-fma3-acc2.c
@@ -3197,7 +3197,7 @@
src/math/sqrt-fma3-nr1fma1adj.c
src/math/sqrt-fma3-nr2fma.c)
-SET(XNNPACK_AVX2_MICROKERNEL_SRCS
+SET(AVX2_MICROKERNEL_SRCS
src/f32-raddexpminusmax/gen/avx2-p5-x64-acc2.c
src/f32-raddexpminusmax/gen/avx2-p5-x64-acc4.c
src/f32-raddexpminusmax/gen/avx2-p5-x64.c
@@ -3441,7 +3441,7 @@
src/qu8-vaddc/gen/minmax-avx2-mul32-ld64-x8.c
src/qu8-vaddc/gen/minmax-avx2-mul32-ld64-x16.c)
-SET(XNNPACK_AVX512F_MICROKERNEL_SRCS
+SET(AVX512F_MICROKERNEL_SRCS
src/f32-dwconv/gen/up16x4-minmax-avx512f-acc2.c
src/f32-dwconv/gen/up16x4-minmax-avx512f.c
src/f32-dwconv/gen/up16x9-minmax-avx512f-acc2.c
@@ -3699,7 +3699,7 @@
src/math/sqrt-avx512f-nr1fma1adj.c
src/math/sqrt-avx512f-nr2fma.c)
-SET(XNNPACK_AVX512SKX_MICROKERNEL_SRCS
+SET(AVX512SKX_MICROKERNEL_SRCS
src/qc8-dwconv/gen/up16x9-minmax-fp32-avx512skx-mul32.c
src/qc8-dwconv/gen/up16x25-minmax-fp32-avx512skx-mul32.c
src/qc8-dwconv/gen/up32x9-minmax-fp32-avx512skx-mul32.c
@@ -3757,7 +3757,7 @@
src/qu8-vaddc/gen/minmax-avx512skx-mul32-ld128-x16.c
src/qu8-vaddc/gen/minmax-avx512skx-mul32-ld128-x32.c)
-SET(XNNPACK_AARCH32_ASM_MICROKERNEL_SRCS
+SET(AARCH32_ASM_MICROKERNEL_SRCS
src/f32-gemm/4x4-aarch32-vfp-ld64.S
src/f32-gemm/4x4-minmax-aarch32-vfp-ld64.S
src/f32-gemm/4x8-minmax-aarch32-neon-cortex-a53.S
@@ -3773,7 +3773,7 @@
src/f32-igemm/gen/4x8-minmax-aarch32-neon-ld64.S
src/f32-igemm/gen/4x8-minmax-aarch32-neon-pld-cortex-a75.S)
-SET(XNNPACK_AARCH64_ASM_MICROKERNEL_SRCS
+SET(AARCH64_ASM_MICROKERNEL_SRCS
src/f16-gemm/gen-inc/1x8inc-minmax-aarch64-neonfp16arith-ld64.S
src/f16-gemm/gen-inc/1x16inc-minmax-aarch64-neonfp16arith-ld32.S
src/f16-gemm/gen-inc/4x8inc-minmax-aarch64-neonfp16arith-ld64.S
@@ -3985,48 +3985,48 @@
src/qs8-igemm/gen/4x16c4-minmax-rndnu-aarch64-neondot-ld64.S
src/qs8-igemm/gen/4x16c4-minmax-rndnu-aarch64-neondot-ld128.S)
-SET(XNNPACK_MICROKERNEL_SRCS ${XNNPACK_SCALAR_MICROKERNEL_SRCS})
+SET(XNNPACK_MICROKERNEL_SRCS ${SCALAR_MICROKERNEL_SRCS})
IF(CMAKE_SYSTEM_PROCESSOR MATCHES "^armv[5-8]" OR IOS_ARCH MATCHES "^armv7")
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEON_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONFMA_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONV8_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEON_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONFMA_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONV8_MICROKERNEL_SRCS})
IF(NOT IOS)
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONDOT_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONDOT_MICROKERNEL_SRCS})
ENDIF()
IF(XNNPACK_ENABLE_ASSEMBLY)
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AARCH32_ASM_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AARCH32_ASM_MICROKERNEL_SRCS})
ENDIF()
ENDIF()
IF(XNNPACK_TARGET_PROCESSOR MATCHES "^(aarch64|arm64)$" OR IOS_ARCH MATCHES "^arm64.*")
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEON_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONFMA_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONV8_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AARCH64_NEONFMA_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_NEONDOT_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEON_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONFMA_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONV8_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AARCH64_NEONFMA_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${NEONDOT_MICROKERNEL_SRCS})
IF(XNNPACK_ENABLE_ASSEMBLY)
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AARCH64_ASM_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AARCH64_ASM_MICROKERNEL_SRCS})
ENDIF()
ENDIF()
IF(XNNPACK_TARGET_PROCESSOR MATCHES "^(i[3-6]86|x86_64|AMD64)$" OR IOS_ARCH MATCHES "^(i386|x86_64|AMD64)$")
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_SSE_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_SSE2_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_SSSE3_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_SSE41_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AVX_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_XOP_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_FMA3_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AVX2_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AVX512F_MICROKERNEL_SRCS})
- LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XNNPACK_AVX512SKX_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${SSE_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${SSE2_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${SSSE3_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${SSE41_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AVX_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${XOP_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${FMA3_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AVX2_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AVX512F_MICROKERNEL_SRCS})
+ LIST(APPEND XNNPACK_MICROKERNEL_SRCS ${AVX512SKX_MICROKERNEL_SRCS})
ENDIF()
IF(XNNPACK_LIBRARY_TYPE STREQUAL "default")
- ADD_LIBRARY(XNNPACK ${XNNPACK_COLD_SRCS} ${XNNPACK_HOT_SRCS} ${XNNPACK_TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
+ ADD_LIBRARY(XNNPACK ${COLD_SRCS} ${HOT_SRCS} ${TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
ELSEIF(XNNPACK_LIBRARY_TYPE STREQUAL "shared")
- ADD_LIBRARY(XNNPACK SHARED ${XNNPACK_COLD_SRCS} ${XNNPACK_HOT_SRCS} ${XNNPACK_TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
+ ADD_LIBRARY(XNNPACK SHARED ${COLD_SRCS} ${HOT_SRCS} ${TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
ELSEIF(XNNPACK_LIBRARY_TYPE STREQUAL "static")
- ADD_LIBRARY(XNNPACK STATIC ${XNNPACK_COLD_SRCS} ${XNNPACK_HOT_SRCS} ${XNNPACK_TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
+ ADD_LIBRARY(XNNPACK STATIC ${COLD_SRCS} ${HOT_SRCS} ${TABLE_SRCS} ${XNNPACK_MICROKERNEL_SRCS})
ELSE()
MESSAGE(FATAL_ERROR "Unsupported XNNPACK library type \"${XNNPACK_LIBRARY_TYPE}\". Must be \"static\", \"shared\", or \"default\"")
ENDIF()
@@ -4035,64 +4035,64 @@
C_EXTENSIONS YES)
IF(CMAKE_SYSTEM_PROCESSOR MATCHES "^armv[5-8]" OR IOS_ARCH MATCHES "^armv7")
SET_PROPERTY(SOURCE ${XNNPACK_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -marm ")
- SET_PROPERTY(SOURCE ${XNNPACK_NEON_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon ")
- SET_PROPERTY(SOURCE ${XNNPACK_NEONFMA_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon-vfpv4 ")
+ SET_PROPERTY(SOURCE ${NEON_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon ")
+ SET_PROPERTY(SOURCE ${NEONFMA_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv7-a -mfpu=neon-vfpv4 ")
IF(IOS)
- SET_PROPERTY(SOURCE ${XNNPACK_NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mcpu=cyclone -mtune=generic ")
- SET_PROPERTY(SOURCE ${XNNPACK_AARCH32_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch ${IOS_ARCH} ")
+ SET_PROPERTY(SOURCE ${NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mcpu=cyclone -mtune=generic ")
+ SET_PROPERTY(SOURCE ${AARCH32_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch ${IOS_ARCH} ")
ELSE()
- SET_PROPERTY(SOURCE ${XNNPACK_NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8-a -mfpu=neon-fp-armv8 ")
- SET_PROPERTY(SOURCE ${XNNPACK_NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+dotprod -mfpu=neon-fp-armv8 ")
+ SET_PROPERTY(SOURCE ${NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8-a -mfpu=neon-fp-armv8 ")
+ SET_PROPERTY(SOURCE ${NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+dotprod -mfpu=neon-fp-armv8 ")
ENDIF()
# Workground the neon detection bug in ARM v8
# Related links:
# https://github.com/android/ndk/issues/910
# https://reviews.llvm.org/D58477
IF(ANDROID_NDK_MAJOR AND ANDROID_NDK_MAJOR LESS 21)
- SET_PROPERTY(SOURCE ${XNNPACK_NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfloat-abi=softfp ")
- SET_PROPERTY(SOURCE ${XNNPACK_NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfloat-abi=softfp ")
+ SET_PROPERTY(SOURCE ${NEONV8_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfloat-abi=softfp ")
+ SET_PROPERTY(SOURCE ${NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfloat-abi=softfp ")
ENDIF()
ENDIF()
IF(XNNPACK_TARGET_PROCESSOR MATCHES "^(aarch64|arm64)$" OR IOS_ARCH MATCHES "^arm64.*")
- SET_PROPERTY(SOURCE ${XNNPACK_AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+fp16 ")
- SET_PROPERTY(SOURCE ${XNNPACK_NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+dotprod ")
- SET_PROPERTY(SOURCE ${XNNPACK_AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+fp16+dotprod ")
+ SET_PROPERTY(SOURCE ${AARCH64_NEONFP16ARITH_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+fp16 ")
+ SET_PROPERTY(SOURCE ${NEONDOT_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+dotprod ")
+ SET_PROPERTY(SOURCE ${AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -march=armv8.2-a+fp16+dotprod ")
IF(IOS)
- SET_PROPERTY(SOURCE ${XNNPACK_AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch ${IOS_ARCH} ")
+ SET_PROPERTY(SOURCE ${AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch ${IOS_ARCH} ")
ENDIF()
IF(XNNPACK_TARGET_PROCESSOR STREQUAL "arm64" AND CMAKE_HOST_SYSTEM_PROCESSOR STREQUAL "x86_64")
- SET_PROPERTY(SOURCE ${XNNPACK_AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch arm64 ")
+ SET_PROPERTY(SOURCE ${AARCH64_ASM_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -arch arm64 ")
ENDIF()
ENDIF()
IF(XNNPACK_TARGET_PROCESSOR MATCHES "^(i[3-6]86|x86|x86_64|AMD64)$" OR IOS_ARCH MATCHES "^(i386|x86_64|AMD64)$")
IF(MSVC)
IF(CMAKE_SYSTEM_PROCESSOR STREQUAL "x86" OR CMAKE_SIZEOF_VOID_P EQUAL 4)
- SET_PROPERTY(SOURCE ${XNNPACK_SSE_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSE2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSSE3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSE41_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
+ SET_PROPERTY(SOURCE ${SSE_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE ")
+ SET_PROPERTY(SOURCE ${SSE2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
+ SET_PROPERTY(SOURCE ${SSSE3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
+ SET_PROPERTY(SOURCE ${SSE41_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:SSE2 ")
ENDIF()
- SET_PROPERTY(SOURCE ${XNNPACK_AVX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
- SET_PROPERTY(SOURCE ${XNNPACK_XOP_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
- SET_PROPERTY(SOURCE ${XNNPACK_FMA3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX512 ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512SKX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX512 ")
+ SET_PROPERTY(SOURCE ${AVX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
+ SET_PROPERTY(SOURCE ${XOP_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
+ SET_PROPERTY(SOURCE ${FMA3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX ")
+ SET_PROPERTY(SOURCE ${AVX2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX2 ")
+ SET_PROPERTY(SOURCE ${AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX512 ")
+ SET_PROPERTY(SOURCE ${AVX512SKX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /arch:AVX512 ")
ELSE()
- SET_PROPERTY(SOURCE ${XNNPACK_SSE_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSE2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSSE3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mssse3 ")
- SET_PROPERTY(SOURCE ${XNNPACK_SSE41_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse4.1 ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mavx ")
- SET_PROPERTY(SOURCE ${XNNPACK_XOP_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mxop ")
- SET_PROPERTY(SOURCE ${XNNPACK_FMA3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfma ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfma -mavx2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mavx512f ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512SKX_MICROKERNEL_SRCS} APPEND_STRIDE PROPERTY COMPILE_FLAGS " -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl ")
+ SET_PROPERTY(SOURCE ${SSE_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse ")
+ SET_PROPERTY(SOURCE ${SSE2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse2 ")
+ SET_PROPERTY(SOURCE ${SSSE3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mssse3 ")
+ SET_PROPERTY(SOURCE ${SSE41_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -msse4.1 ")
+ SET_PROPERTY(SOURCE ${AVX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mavx ")
+ SET_PROPERTY(SOURCE ${XOP_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mxop ")
+ SET_PROPERTY(SOURCE ${FMA3_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfma ")
+ SET_PROPERTY(SOURCE ${AVX2_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mfma -mavx2 ")
+ SET_PROPERTY(SOURCE ${AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -mavx512f ")
+ SET_PROPERTY(SOURCE ${AVX512SKX_MICROKERNEL_SRCS} APPEND_STRIDE PROPERTY COMPILE_FLAGS " -mavx512f -mavx512cd -mavx512bw -mavx512dq -mavx512vl ")
IF(MINGW OR CMAKE_SYSTEM_NAME MATCHES "^(CYGWIN|MSYS)$")
# Work-around for https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65782
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -fno-asynchronous-unwind-tables ")
- SET_PROPERTY(SOURCE ${XNNPACK_AVX512SKX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -fno-asynchronous-unwind-tables ")
+ SET_PROPERTY(SOURCE ${AVX512F_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -fno-asynchronous-unwind-tables ")
+ SET_PROPERTY(SOURCE ${AVX512SKX_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -fno-asynchronous-unwind-tables ")
ENDIF()
ENDIF()
ENDIF()
@@ -4103,25 +4103,25 @@
IF(${CMAKE_VERSION} VERSION_LESS "3.8.0")
IF(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
SET_PROPERTY(SOURCE ${XNNPACK_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /O2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /O2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /O1 ")
+ SET_PROPERTY(SOURCE ${HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /O2 ")
+ SET_PROPERTY(SOURCE ${COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " /O1 ")
ENDIF()
ELSE()
SET_PROPERTY(SOURCE ${XNNPACK_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: /O2 >")
- SET_PROPERTY(SOURCE ${XNNPACK_HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: /O2 >")
- SET_PROPERTY(SOURCE ${XNNPACK_COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: /O1 >")
+ SET_PROPERTY(SOURCE ${HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: /O2 >")
+ SET_PROPERTY(SOURCE ${COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: /O1 >")
ENDIF()
ELSE()
IF(${CMAKE_VERSION} VERSION_LESS "3.8.0")
IF(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
SET_PROPERTY(SOURCE ${XNNPACK_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -O2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -O2 ")
- SET_PROPERTY(SOURCE ${XNNPACK_COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -Os ")
+ SET_PROPERTY(SOURCE ${HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -O2 ")
+ SET_PROPERTY(SOURCE ${COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS " -Os ")
ENDIF()
ELSE()
SET_PROPERTY(SOURCE ${XNNPACK_MICROKERNEL_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: -O2 >")
- SET_PROPERTY(SOURCE ${XNNPACK_HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: -O2 >")
- SET_PROPERTY(SOURCE ${XNNPACK_COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: -Os >")
+ SET_PROPERTY(SOURCE ${HOT_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: -O2 >")
+ SET_PROPERTY(SOURCE ${COLD_SRCS} APPEND_STRING PROPERTY COMPILE_FLAGS "$<$<NOT:$<CONFIG:Debug>>: -Os >")
ENDIF()
ENDIF()