Update XNNPACK to Branch with Kleidi QB4

pytorch · Sep 14, 2024 · 61a8270 · 61a8270
1 parent 2001b3c
commit 61a8270
Show file tree

Hide file tree

Showing 9 changed files with 160 additions and 1,183 deletions.
diff --git a/backends/xnnpack/cmake/Dependencies.cmake b/backends/xnnpack/cmake/Dependencies.cmake
@@ -43,6 +43,10 @@ set(XNNPACK_ENABLE_KLEIDIAI
 add_subdirectory("${XNNPACK_SOURCE_DIR}")
 include_directories(SYSTEM ${XNNPACK_INCLUDE_DIR})
 list(APPEND xnnpack_third_party XNNPACK)
+install(TARGETS microkernels-prod
+    LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    PUBLIC_HEADER DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})
 
 # Revert PIC Flag to what it originally was
 set(CMAKE_POSITION_INDEPENDENT_CODE

diff --git a/backends/xnnpack/third-party/XNNPACK b/backends/xnnpack/third-party/XNNPACK
diff --git a/backends/xnnpack/third-party/cpuinfo b/backends/xnnpack/third-party/cpuinfo
diff --git a/backends/xnnpack/third-party/generate-xnnpack-wrappers.py b/backends/xnnpack/third-party/generate-xnnpack-wrappers.py
@@ -8,88 +8,13 @@
 import logging
 
 BANNER = "Auto-generated by generate-wrappers.py script. Do not modify"
-WRAPPER_SRC_NAMES = {
-    "PROD_SCALAR_MICROKERNEL_SRCS": None,
-    "PROD_FMA_MICROKERNEL_SRCS": "defined(__riscv) || defined(__riscv__)",
-    "PROD_ARMSIMD32_MICROKERNEL_SRCS": "defined(__arm__)",
-    "PROD_FP16ARITH_MICROKERNEL_SRCS": "defined(__arm__)",
-    "PROD_NEON_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONFP16_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONFMA_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEON_AARCH64_MICROKERNEL_SRCS": "defined(__aarch64__)",
-    "PROD_NEONV8_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONFP16ARITH_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONFP16ARITH_AARCH64_MICROKERNEL_SRCS": "defined(__aarch64__)",
-    "PROD_NEONDOT_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONDOT_AARCH64_MICROKERNEL_SRCS": "defined(__aarch64__)",
-    "PROD_NEONDOTFP16ARITH_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_NEONDOTFP16ARITH_AARCH64_MICROKERNEL_SRCS": "defined(__aarch64__)",
-    "PROD_NEONI8MM_MICROKERNEL_SRCS": "defined(__arm__) || defined(__aarch64__)",
-    "PROD_SSE_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_SSE2_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_SSSE3_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_SSE41_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_F16C_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_XOP_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_FMA3_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX2_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX512F_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX512SKX_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX512VBMI_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX512VNNI_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_AVX512VNNIGFNI_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "PROD_RVV_MICROKERNEL_SRCS": "defined(__riscv) || defined(__riscv__)",
-    "PROD_AVXVNNI_MICROKERNEL_SRCS": "defined(__i386__) || defined(__i686__) || defined(__x86_64__)",
-    "AARCH32_ASM_MICROKERNEL_SRCS": "defined(__arm__)",
-    "AARCH64_ASM_MICROKERNEL_SRCS": "defined(__aarch64__)",
-
-    # add non-prod microkernel sources here:
-}
 
 SRC_NAMES = {
     "OPERATOR_SRCS",
     "SUBGRAPH_SRCS",
     "LOGGING_SRCS",
     "XNNPACK_SRCS",
     "TABLE_SRCS",
-    "JIT_SRCS",
-    "PROD_SCALAR_MICROKERNEL_SRCS",
-    "PROD_FMA_MICROKERNEL_SRCS",
-    "PROD_ARMSIMD32_MICROKERNEL_SRCS",
-    "PROD_FP16ARITH_MICROKERNEL_SRCS",
-    "PROD_NEON_MICROKERNEL_SRCS",
-    "PROD_NEONFP16_MICROKERNEL_SRCS",
-    "PROD_NEONFMA_MICROKERNEL_SRCS",
-    "PROD_NEON_AARCH64_MICROKERNEL_SRCS",
-    "PROD_NEONV8_MICROKERNEL_SRCS",
-    "PROD_NEONFP16ARITH_MICROKERNEL_SRCS",
-    "PROD_NEONFP16ARITH_AARCH64_MICROKERNEL_SRCS",
-    "PROD_NEONDOT_MICROKERNEL_SRCS",
-    "PROD_NEONDOT_AARCH64_MICROKERNEL_SRCS",
-    "PROD_NEONDOTFP16ARITH_MICROKERNEL_SRCS",
-    "PROD_NEONDOTFP16ARITH_AARCH64_MICROKERNEL_SRCS",
-    "PROD_NEONI8MM_MICROKERNEL_SRCS",
-    "PROD_SSE_MICROKERNEL_SRCS",
-    "PROD_SSE2_MICROKERNEL_SRCS",
-    "PROD_SSSE3_MICROKERNEL_SRCS",
-    "PROD_SSE41_MICROKERNEL_SRCS",
-    "PROD_AVX_MICROKERNEL_SRCS",
-    "PROD_F16C_MICROKERNEL_SRCS",
-    "PROD_XOP_MICROKERNEL_SRCS",
-    "PROD_FMA3_MICROKERNEL_SRCS",
-    "PROD_AVX2_MICROKERNEL_SRCS",
-    "PROD_AVX512F_MICROKERNEL_SRCS",
-    "PROD_AVX512SKX_MICROKERNEL_SRCS",
-    "PROD_AVX512VBMI_MICROKERNEL_SRCS",
-    "PROD_AVX512VNNI_MICROKERNEL_SRCS",
-    "PROD_AVX512VNNIGFNI_MICROKERNEL_SRCS",
-    "PROD_RVV_MICROKERNEL_SRCS",
-    "PROD_AVXVNNI_MICROKERNEL_SRCS",
-    "AARCH32_ASM_MICROKERNEL_SRCS",
-    "AARCH64_ASM_MICROKERNEL_SRCS",
-
-    # add non-prod microkernel sources here:
 }
 
 def handle_singleline_parse(line):
+2 −2		.github/workflows/build.yml
+1 −1		CMakeLists.txt
+3 −1		README.md
+9 −0		include/cpuinfo.h
+59 −0		scripts/android-riscv64-build.sh
+3 −0		src/arm/linux/aarch64-isa.c
+1 −0		src/arm/linux/api.h
+11 −15		src/freebsd/topology.c
+21 −5		src/x86/freebsd/init.c