cpu: add loongson architecture support

CMakeLists.txt

@@ -95,6 +95,8 @@ if(NOT DNNL_TARGET_ARCH)
         set(DNNL_TARGET_ARCH "S390X")
     elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(rv.*|RV.*|riscv.*|RISCV.*)")
         set(DNNL_TARGET_ARCH "RV64")
+    elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(loongarch64.*|LOONGARCH64.*)")
+        set(DNNL_TARGET_ARCH "LOONGARCH64")
     else()
         set(DNNL_TARGET_ARCH "X64")
     endif()

README.md

@@ -74,10 +74,11 @@ oneDNN supports platforms based on the following architectures:
 - [OpenPOWER](https://openpowerfoundation.org/) / [IBM Power ISA](https://en.wikipedia.org/wiki/Power_ISA).
 - [IBMz z/Architecture (s390x)](https://en.wikipedia.org/wiki/Z/Architecture).
 - [RISC-V 64-bit (RV64)](https://en.wikipedia.org/wiki/RISC-V).
+- [LOONGARCH 64 bit](https://docs.kernel.org/arch/loongarch/introduction.html).
 
 > **WARNING**
 >
-> Power ISA (PPC64), IBMz (s390x), and RISC-V (RV64) support is
+> Power ISA (PPC64), IBMz (s390x), RISC-V (RV64) and Loongson (LOONGARCH64) support is
 > **experimental** with limited testing validation.
 
 The library is optimized for the following CPUs:

cmake/platform.cmake

@@ -363,6 +363,8 @@ elseif(UNIX OR MINGW)
         elseif(DNNL_TARGET_ARCH STREQUAL "RV64")
             # G = General-purpose extensions, C = Compression extension (very common).
             append(DEF_ARCH_OPT_FLAGS "-march=rv64gc")
+	elseif(DNNL_TARGET_ARCH STREQUAL "LOONGARCH64")
+            append(DEF_ARCH_OPT_FLAGS "-march=loongarch64")
         elseif(DNNL_TARGET_ARCH STREQUAL "X64")
             platform_gnu_x64_arch_ccxx_flags(DEF_ARCH_OPT_FLAGS)
         endif()

src/cpu/README.md

@@ -46,6 +46,7 @@ enable or disable parts of code. There the following macros defined:
 - `DNNL_PPC64` is 1 on OpenPOWER / IBM Power architecture;
 - `DNNL_S390X` is 1 on IBMz / s390x architecture;
 - `DNNL_RV64` is 1 on RISC-V architecture;
+- `DNNL_LOONGARCH64` is 1 on Loongson LOONGARCH64 architecture;
 - `DNNL_ARCH_GENERIC` is 1 on other platforms.
 Only one of the macros above is defined to 1. All others are defined to 0.
 

src/cpu/platform.hpp

@@ -30,12 +30,13 @@
 // - DNNL_PPC64
 // - DNNL_S390X
 // - DNNL_RV64
+// - DNNL_LOONGARCH64
 // - DNNL_ARCH_GENERIC
 // Target architecture macro is set to 1, others to 0. All macros are defined.
 
 #if defined(DNNL_X64) + defined(DNNL_AARCH64) + defined(DNNL_PPC64) \
                 + defined(DNNL_S390X) + defined(DNNL_RV64) \
-                + defined(DNNL_ARCH_GENERIC) \
+                + defined(DNNL_LOONGARCH64) + defined(DNNL_ARCH_GENERIC) \
         == 0
 #if defined(__x86_64__) || defined(_M_X64)
 #define DNNL_X64 1
@@ -47,14 +48,16 @@
 #define DNNL_S390X 1
 #elif defined(__riscv)
 #define DNNL_RV64 1
+#elif defined(__loongarch64)
+#define DNNL_LOONGARCH64 1
 #else
 #define DNNL_ARCH_GENERIC 1
 #endif
 #endif // defined(DNNL_X64) + ... == 0
 
 #if defined(DNNL_X64) + defined(DNNL_AARCH64) + defined(DNNL_PPC64) \
                 + defined(DNNL_S390X) + defined(DNNL_RV64) \
-                + defined(DNNL_ARCH_GENERIC) \
+                + defined(DNNL_LOONGARCH64) + defined(DNNL_ARCH_GENERIC) \
         != 1
 #error One and only one architecture should be defined at a time
 #endif
@@ -74,6 +77,9 @@
 #if !defined(DNNL_RV64)
 #define DNNL_RV64 0
 #endif
+#if !defined(DNNL_LOONGARCH64)
+#define DNNL_LOONGARCH64 0
+#endif
 #if !defined(DNNL_ARCH_GENERIC)
 #define DNNL_ARCH_GENERIC 0
 #endif