Bug fixes from power9 compilation

Signed-off-by: WillTrojak <[email protected]>
WillTrojak · Sep 19, 2024 · f291779 · f291779
1 parent e072f53
commit f291779
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 13 deletions.
diff --git a/src/generator_gemm_mma_microkernel.c b/src/generator_gemm_mma_microkernel.c
@@ -51,7 +51,7 @@ void libxsmm_generator_gemm_mma_mk_load_trans( libxsmm_generated_code   *io_gene
   unsigned int l_vec_ele = l_vec_len / l_databytes;
   unsigned int l_m_blocks = i_m / l_vec_ele;
   unsigned int l_n_blocks = (i_n + l_vec_ele - 1) / l_vec_ele;
-  unsigned int l_m_part;
+  unsigned int l_m_part = -1;
   unsigned int l_packed = ( ( i_m % l_vec_ele ) == 0 ) ? 1 : 0;
 
   /* Local copy of pointer */
@@ -232,7 +232,7 @@ void libxsmm_generator_gemm_mma_mk_load_acc( libxsmm_generated_code *io_generate
   unsigned int l_vec_ele = l_vec_len / l_databytes;
   unsigned int l_acc_vec = LIBXSMM_PPC64LE_ACC_WIDTH / LIBXSMM_PPC64LE_VSR_WIDTH;
   unsigned int l_m_blocks = i_m / l_vec_ele;
-  unsigned int l_m_part;
+  unsigned int l_m_part = -1;
   unsigned int l_packed = ( ( i_m % l_vec_ele ) == 0 ) ? 1 : 0;
 
   /* Local copies of pointer */
@@ -370,7 +370,7 @@ void libxsmm_generator_gemm_mma_mk_store_acc( libxsmm_generated_code *io_generat
   unsigned int l_vec_ele = l_vec_len / l_databytes;
   unsigned int l_acc_vec = LIBXSMM_PPC64LE_ACC_WIDTH / LIBXSMM_PPC64LE_VSR_WIDTH;
   unsigned int l_m_blocks = i_m / l_vec_ele;
-  unsigned int l_m_part;
+  unsigned int l_m_part = -1;
   unsigned int l_packed = ( ( i_m % l_vec_ele ) == 0 ) ? 1 : 0;
 
   /* Local copies of pointer */

diff --git a/src/libxsmm_cpuid_ppc.c b/src/libxsmm_cpuid_ppc.c
@@ -35,15 +35,14 @@ LIBXSMM_API int libxsmm_cpuid_ppc(libxsmm_cpuid_info* info)
 # if defined(LIBXSMM_CPUID_PPC_BASELINE)
     result = LIBXSMM_CPUID_PPC_BASELINE;
 # else
-    if (__builtin_cpu_supports("mma") > 0) {
-      result = LIBXSMM_PPC64LE_MMA;
-    }
-    else if (__builtin_cpu_supports("vsx") > 0) {
-      result = LIBXSMM_PPC64LE_VSX;
-    }
-    else if (__builtin_cpu_supports("fpu") > 0) {
-      result = LIBXSMM_PPC64LE_FPF;
-    }
+
+#  if defined(_ARCH_PWR10)
+    result = LIBXSMM_PPC64LE_MMA;
+#  elif defined(_ARCH_PWR9)
+    result = LIBXSMM_PPC64LE_VSX;
+#  elif defined(_ARCH_PWR8)
+    result = LIBXSMM_PPC64LE_FPF;
+#  endif
 # endif
   }
 

diff --git a/version.txt b/version.txt
@@ -1 +1 @@
-feature/power-1.17-3771
+feature/power-1.17-3772