Revert "[aarch64] Implement QGEMM kernels with UMMLA/SMMLA instructio…

…ns (#17160)" This reverts commit 780ee18.
microsoft · Oct 25, 2023 · b2f07a6 · b2f07a6
1 parent 76e275b
commit b2f07a6
Show file tree

Hide file tree

Showing 9 changed files with 0 additions and 3,833 deletions.
diff --git a/cmake/onnxruntime_mlas.cmake b/cmake/onnxruntime_mlas.cmake
@@ -325,9 +325,7 @@ else()
           ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelUdot.S
-          ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelUmmla.S
           ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelSdot.S
-          ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelSmmla.S
           ${MLAS_SRC_DIR}/aarch64/SgemmKernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/SgemvKernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/SymQgemmS8KernelNeon.S
@@ -336,8 +334,6 @@ else()
           ${MLAS_SRC_DIR}/qgemm_kernel_neon.cpp
           ${MLAS_SRC_DIR}/qgemm_kernel_udot.cpp
           ${MLAS_SRC_DIR}/qgemm_kernel_sdot.cpp
-          ${MLAS_SRC_DIR}/qgemm_kernel_ummla.cpp
-          ${MLAS_SRC_DIR}/qgemm_kernel_smmla.cpp
         )
         if (NOT APPLE)
           set(mlas_platform_srcs
@@ -352,8 +348,6 @@ else()
           set_source_files_properties(${MLAS_SRC_DIR}/activate_fp16.cpp PROPERTIES COMPILE_FLAGS " -march=armv8.2-a+fp16 ")
           set_source_files_properties(${MLAS_SRC_DIR}/dwconv.cpp PROPERTIES COMPILE_FLAGS " -march=armv8.2-a+fp16 ")
           set_source_files_properties(${MLAS_SRC_DIR}/pooling_fp16.cpp PROPERTIES COMPILE_FLAGS " -march=armv8.2-a+fp16 ")
-          set_source_files_properties(${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelUmmla.S PROPERTIES COMPILE_FLAGS " -march=armv8.2-a+i8mm ")
-          set_source_files_properties(${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelSmmla.S PROPERTIES COMPILE_FLAGS " -march=armv8.2-a+i8mm ")
         endif()
 
         if(ONNXRUNTIME_MLAS_MULTI_ARCH)

diff --git a/onnxruntime/core/common/cpuid_info.cc b/onnxruntime/core/common/cpuid_info.cc
@@ -22,14 +22,6 @@
 #define HWCAP_ASIMDDP (1 << 20)
 #endif
 
-#ifndef HWCAP2_I8MM
-#define HWCAP2_I8MM (1 << 13)
-#endif
-
-#ifndef HWCAP2_SVEI8MM
-#define HWCAP2_SVEI8MM (1 << 9)
-#endif
-
 #endif  // ARM
 
 #endif  // Linux
@@ -146,9 +138,6 @@ void CPUIDInfo::ArmLinuxInit() {
   is_hybrid_ = cpuinfo_get_uarchs_count() > 1;
   has_arm_neon_dot_ = cpuinfo_has_arm_neon_dot();
   has_fp16_ = cpuinfo_has_arm_neon_fp16_arith();
-  has_arm_neon_i8mm_ = cpuinfo_has_arm_i8mm();
-  has_arm_sve_i8mm_ = cpuinfo_has_arm_sve() && cpuinfo_has_arm_i8mm();
-
   const uint32_t core_cnt = cpuinfo_get_cores_count();
   core_uarchs_.resize(core_cnt, cpuinfo_uarch_unknown);
   is_armv8_narrow_ld_.resize(core_cnt, false);
@@ -173,10 +162,6 @@ void CPUIDInfo::ArmLinuxInit() {
   pytorch_cpuinfo_init_ = false;
   has_arm_neon_dot_ = ((getauxval(AT_HWCAP) & HWCAP_ASIMDDP) != 0);
   has_fp16_ |= has_arm_neon_dot_;
-
-  has_arm_neon_i8mm_ = ((getauxval(AT_HWCAP2) & HWCAP2_I8MM) != 0);
-  has_arm_sve_i8mm_ = ((getauxval(AT_HWCAP2) & HWCAP2_SVEI8MM) != 0);
-
 #endif
 }
 
@@ -271,9 +256,6 @@ void CPUIDInfo::ArmWindowsInit() {
 
   has_arm_neon_dot_ = (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE) != 0);
   has_fp16_ |= has_arm_neon_dot_;
-  /* TODO: implement them when hw+sw is available for testing these features */
-  has_arm_neon_i8mm_ = false;
-  has_arm_sve_i8mm_ = false;
 }
 
 #endif /* (arm or arm64) and windows */

diff --git a/onnxruntime/core/common/cpuid_info.h b/onnxruntime/core/common/cpuid_info.h
@@ -28,8 +28,6 @@ class CPUIDInfo {
 
   // ARM
   bool HasArmNeonDot() const { return has_arm_neon_dot_; }
-  bool HasArmNeon_I8MM() const { return has_arm_neon_i8mm_; }
-  bool HasArmSVE_I8MM() const { return has_arm_sve_i8mm_; }
 
   uint32_t GetCurrentCoreIdx() const;
 
@@ -123,8 +121,6 @@ class CPUIDInfo {
 
   bool has_arm_neon_dot_{false};
   bool has_fp16_{false};
-  bool has_arm_neon_i8mm_{false};
-  bool has_arm_sve_i8mm_{false};
 
 #ifdef CPUIDINFO_ARCH_X86