Modified the cmake logic

Signed-off-by: divya2108 <divya.kotadiya@fujitsu.com>
dmlc · Oct 8, 2024 · 8f5a9c6 · 8f5a9c6
1 parent c8717c1
commit 8f5a9c6
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 46 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -38,6 +38,12 @@ elseif(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
   endif()
 endif()
 
+include(${xgboost_SOURCE_DIR}/cmake/CheckSVEsupport.cmake)
+check_xgboost_sve_support()
+if(XGBOOST_COMPILER_HAS_ARM_SVE)
+    add_compile_definitions(XGBOOST_SVE_COMPILER_SUPPORT)
+endif()
+
 include(${xgboost_SOURCE_DIR}/cmake/PrefetchIntrinsics.cmake)
 find_prefetch_intrinsics()
 include(${xgboost_SOURCE_DIR}/cmake/Version.cmake)
@@ -348,13 +354,6 @@ target_include_directories(xgboost
   $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}/include>)
 #-- End shared library
 
-include(${xgboost_SOURCE_DIR}/cmake/CheckSVEsupport.cmake)
-check_xgboost_sve_support()
-if(XGBOOST_ARM_SVE_HARDWARE_SUPPORT)
-    target_compile_definitions(objxgboost PUBLIC XGBOOST_SVE_SUPPORT_DETECTED)
-    target_compile_options(objxgboost PRIVATE ${XGBOOST_SVE_FLAGS})
-endif()
-
 #-- CLI for xgboost
 if(BUILD_DEPRECATED_CLI)
   add_executable(runxgboost ${xgboost_SOURCE_DIR}/src/cli_main.cc)

diff --git a/cmake/CheckSVEsupport.cmake b/cmake/CheckSVEsupport.cmake
@@ -20,30 +20,6 @@ if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
 
     if(XGBOOST_COMPILER_HAS_ARM_SVE)
         message(STATUS "ARM SVE compiler support detected")
-
-        # Check for hardware support
-        set(SOURCE_CODE "
-        #include <sys/prctl.h>
-        int main() {
-            int ret = prctl(PR_SVE_GET_VL);
-            return ret >= 0 ? 0 : 1;
-        }
-        ")
-        file(WRITE ${CMAKE_BINARY_DIR}/check_sve_support.c "${SOURCE_CODE}")
-        try_run(RUN_RESULT COMPILE_RESULT
-            ${CMAKE_BINARY_DIR}/check_sve_support_output
-            ${CMAKE_BINARY_DIR}/check_sve_support.c
-        )
-
-        if(RUN_RESULT EQUAL 0)
-            message(STATUS "ARM SVE hardware support detected")
-            # Apply the SVE flags and definitions specifically to the xgboost target
-            set(XGBOOST_ARM_SVE_HARDWARE_SUPPORT TRUE PARENT_SCOPE)
-            set(XGBOOST_SVE_FLAGS "-march=armv8-a+sve" PARENT_SCOPE)
-            set(XGBOOST_SVE_DEFINITIONS "-DXGBOOST_SVE_SUPPORT_DETECTED" PARENT_SCOPE)
-        else()
-            message(STATUS "ARM SVE hardware support not detected")
-        endif()
     else()
         message(STATUS "ARM SVE compiler support not detected")
     endif()

diff --git a/src/common/hist_util.cc b/src/common/hist_util.cc
@@ -16,7 +16,11 @@
 #include "xgboost/context.h"  // for Context
 #include "xgboost/data.h"     // for SparsePage, SortedCSCPage
 
-#ifdef XGBOOST_SVE_SUPPORT_DETECTED
+#ifdef __linux__
+#include <sys/prctl.h>
+#endif
+
+#ifdef XGBOOST_SVE_COMPILER_SUPPORT
 #include <arm_sve.h>  // to leverage sve intrinsics
 #endif
 
@@ -183,8 +187,9 @@ class GHistBuildingManager {
   }
 };
 
-#ifdef XGBOOST_SVE_SUPPORT_DETECTED
+#ifdef XGBOOST_SVE_COMPILER_SUPPORT
 template <typename BinIdxType>
+__attribute__((target("arch=armv8-a+sve")))
 inline void UpdateHistogramWithSVE(size_t row_size, const BinIdxType *gr_index_local,
                                    const std::uint32_t *offsets, double *hist_data,
                                    const float *p_gpair, size_t idx_gh, const uint32_t two,
@@ -241,6 +246,12 @@ inline void UpdateHistogramWithSVE(size_t row_size, const BinIdxType *gr_index_l
 }
 #endif
 
+// Returns true if SVE ISA is available on the current CPU
+bool check_sve_hw_support() {
+  int ret = prctl(PR_SVE_GET_VL);
+  return ret >= 0 ? 1 : 0;
+}
+
 template <bool do_prefetch, class BuildingManager>
 void RowsWiseBuildHistKernel(Span<GradientPair const> gpair, Span<bst_idx_t const> row_indices,
                              const GHistIndexMatrix &gmat, GHistRow hist) {
@@ -302,20 +313,22 @@ void RowsWiseBuildHistKernel(Span<GradientPair const> gpair, Span<bst_idx_t cons
     }
     const BinIdxType *gr_index_local = gradient_index + icol_start;
 
-#ifdef XGBOOST_SVE_SUPPORT_DETECTED
-    UpdateHistogramWithSVE(row_size, gr_index_local, offsets, hist_data, p_gpair, idx_gh, two,
-                           kAnyMissing);
-#else
-    // The trick with pgh_t buffer helps the compiler to generate faster binary.
-    const float pgh_t[] = {p_gpair[idx_gh], p_gpair[idx_gh + 1]};
-    for (size_t j = 0; j < row_size; ++j) {
-      const uint32_t idx_bin =
-          two * (static_cast<uint32_t>(gr_index_local[j]) + (kAnyMissing ? 0 : offsets[j]));
-      auto hist_local = hist_data + idx_bin;
-      *(hist_local) += pgh_t[0];
-      *(hist_local + 1) += pgh_t[1];
+    if (check_sve_hw_support()) {
+      #ifdef XGBOOST_SVE_COMPILER_SUPPORT
+        UpdateHistogramWithSVE(row_size, gr_index_local, offsets, hist_data, p_gpair, idx_gh, two,
+                                kAnyMissing);
+      #endif
+    } else {
+        // The trick with pgh_t buffer helps the compiler to generate faster binary.
+        const float pgh_t[] = {p_gpair[idx_gh], p_gpair[idx_gh + 1]};
+        for (size_t j = 0; j < row_size; ++j) {
+          const uint32_t idx_bin =
+              two * (static_cast<uint32_t>(gr_index_local[j]) + (kAnyMissing ? 0 : offsets[j]));
+          auto hist_local = hist_data + idx_bin;
+          *(hist_local) += pgh_t[0];
+          *(hist_local + 1) += pgh_t[1];
+        }
     }
-#endif
   }
 }