Merge pull request #105 from xianyi/develop

rebase

Author: martin-frbg

Makefile.system

@@ -641,18 +641,22 @@ DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
 DYNAMIC_CORE += POWER9
 DYNAMIC_CORE += POWER10
+CCOMMON_OPT += -DHAVE_P10_SUPPORT
 endif
 ifeq ($(C_COMPILER), GCC)
 ifeq ($(GCCVERSIONGT5), 1)
 DYNAMIC_CORE += POWER9
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
-ifeq ($(GCCVERSIONGTEQ11), 1)
+LDVERSIONGTEQ35 := $(shell expr `ld --version | head -1 | cut -f2 -d "." | cut -f1 -d "-"` >= 35)
+ifeq ($(GCCVERSIONGTEQ11)$(LDVERSIONGTEQ35), 11)
 DYNAMIC_CORE += POWER10
+CCOMMON_OPT += -DHAVE_P10_SUPPORT
 else ifeq ($(GCCVERSIONGTEQ10), 1)
-ifeq ($(GCCMINORVERSIONGTEQ2), 1)
+ifeq ($(GCCMINORVERSIONGTEQ2)$(LDVERSIONGTEQ35), 11)
 DYNAMIC_CORE += POWER10
+CCOMMON_OPT += -DHAVE_P10_SUPPORT
 endif
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)

cmake/arch.cmake

@@ -49,6 +49,7 @@ if (DYNAMIC_ARCH)
 
   if (POWER)
 	  set(DYNAMIC_CORE POWER6 POWER8 POWER9 POWER10)
+	  set(CCOMMON_OPT "${CCOMMON_OPT} -DHAVE_P10_SUPPORT")
   endif ()
 
   if (X86)

driver/others/dynamic_power.c

@@ -6,10 +6,10 @@ extern gotoblas_t gotoblas_POWER8;
 #if (!defined __GNUC__) || ( __GNUC__ >= 6)
 extern gotoblas_t gotoblas_POWER9;
 #endif
-#if (!defined __GNUC__) || ( __GNUC__ >= 11) \
-     || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
-#define HAVE_P10_SUPPORT 1
-#endif
+//#if (!defined __GNUC__) || ( __GNUC__ >= 11) \
+//     || (__GNUC__ == 10 && __GNUC_MINOR__ >= 2)
+//#define HAVE_P10_SUPPORT 1
+//#endif
 #ifdef HAVE_P10_SUPPORT
 extern gotoblas_t gotoblas_POWER10;
 #endif

kernel/x86_64/caxpy_microk_bulldozer-2.c

@@ -122,7 +122,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
@@ -189,9 +189,10 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);
 

kernel/x86_64/caxpy_microk_haswell-2.c

@@ -120,7 +120,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15", 

kernel/x86_64/caxpy_microk_sandy-2.c

@@ -104,7 +104,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15", 

kernel/x86_64/caxpy_microk_steamroller-2.c

@@ -122,7 +122,7 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
@@ -189,9 +189,10 @@ static void caxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (alpha),  // 4
           "r" (mvec)    // 5
 	: "cc", 
-	  "%xmm0", "%xmm1",
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
 	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);
 

kernel/x86_64/daxpy_microk_haswell-2.c

@@ -67,8 +67,9 @@ static void daxpy_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *alpha)
           "r" (y),      // 3
           "r" (alpha)   // 4
 	: "cc", 
-	  "%xmm0", 
-	  "%xmm8", "%xmm9", "%xmm10", "%xmm11", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7",
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11",
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);

kernel/x86_64/ddot_microk_haswell-2.c

@@ -84,8 +84,9 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
           "r" (y),      // 3
           "r" (dot)     // 4
 	: "cc", 
-	  "%xmm4", "%xmm5", 
-	  "%xmm6", "%xmm7", 
+	  "%xmm0", "%xmm1", "%xmm2", "%xmm3",
+	  "%xmm4", "%xmm5", "%xmm6", "%xmm7",
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11",
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);

kernel/x86_64/ddot_microk_piledriver-2.c

@@ -91,6 +91,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	: "cc", 
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11",
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);
@@ -155,6 +156,7 @@ static void ddot_kernel_8( BLASLONG n, FLOAT *x, FLOAT *y, FLOAT *dot)
 	: "cc", 
 	  "%xmm0", "%xmm1", "%xmm2", "%xmm3", 
 	  "%xmm4", "%xmm5", "%xmm6", "%xmm7", 
+	  "%xmm8", "%xmm9", "%xmm10", "%xmm11",
 	  "%xmm12", "%xmm13", "%xmm14", "%xmm15",
 	  "memory"
 	);