Merge branch 'develop' of https://github.com/OpenMathLib/OpenBLAS into develop

Author: ChipKerchner

Makefile.install

@@ -32,7 +32,7 @@ PKG_EXTRALIB := $(EXTRALIB)
 ifeq ($(INTERFACE64),1)
 	SUFFIX64=64
 endif
-PKGFILE="$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/$(LIBSONAMEBASE)$(SUFFIX64).pc"
+PKGFILE := $(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/$(LIBSONAMEBASE)$(SUFFIX64).pc
 
 ifeq ($(USE_OPENMP), 1)
   ifeq ($(C_COMPILER), PGI)
@@ -191,8 +191,6 @@ endif
 ifeq ($(INTERFACE64),1)
 	SUFFIX64=64
 endif
-	PKGFILE="$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)/$(LIBSONAMEBASE)$(SUFFIX64).pc"
-
 	@echo Generating $(LIBSONAMEBASE)$(SUFFIX64).pc in "$(DESTDIR)$(OPENBLAS_PKGCONFIG_DIR)"
 	@echo 'libdir='$(OPENBLAS_LIBRARY_DIR) > "$(PKGFILE)"
 	@echo 'libprefix='$(LIBNAMEPREFIX) >> "$(PKGFILE)"

Makefile.system

@@ -477,7 +477,7 @@ EXTRALIB	+= -lm
 endif
 
 ifeq ($(OSNAME), $(filter $(OSNAME),FreeBSD OpenBSD NetBSD DragonFly))
-ifeq ($(ARCH), $(filter $(ARCH),arm arm64))
+ifeq ($(ARCH), $(filter $(ARCH),arm arm64 power))
 EXTRALIB	+= -lm
 endif
 endif

interface/trsm.c

@@ -381,8 +381,12 @@ if (strcmp(gotoblas_corename(), "armv9sme") == 0
 #endif
 #endif
 
+
+
+//end of the ifndef CBLAS ... else ... section	
 #endif
 
+  if ((args.m == 0) || (args.n == 0)) return;
   IDEBUG_START;
 
   FUNCTION_PROFILE_START();

kernel/Makefile

@@ -97,11 +97,11 @@ else ifeq ($(TARGET_CORE), ZEN)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(AVX2OPT)
 else ifeq ($(TARGET_CORE), LOONGSON3R4)
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) $(MSA_FLAGS)
-else ifneq ($(filter NEOVERSEN2 NEOVERSEV1, $(TARGET_CORE)),)
+else ifneq ($(filter NEOVERSEN2 NEOVERSEV1 ARMV8SVE A64FX, $(TARGET_CORE)),)
  ifeq ($(C_COMPILER), PGI)
   override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -Msve_intrinsics
  else
-  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)
+  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=armv8.4-a+sve+bf16
  endif
 else
  override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE)

kernel/riscv64/dgemm_kernel_8x8_zvl256b.c

@@ -1628,11 +1628,12 @@ static void NM_TAIL(BLASLONG K, BLASLONG M, const BLASLONG m_edge, const BLASLON
         }
     }
 }
+
 int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT* A, FLOAT* B, FLOAT* C, BLASLONG ldc)
 {
     if (K <= 0) return 0;
     const BLASLONG m_edge = M & 7;
-    const bool S = (M == (ldc & 0x7));
+    const bool S = (ldc == m_edge);
 
     // -- MAIN PASS
 
@@ -1689,24 +1690,6 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT* A, FLOAT* B, F
                 resultEF = __riscv_vfmacc_vf_f64m2( resultEF, B7, A00, 8 );
             }
 
-            // LMUL = 2 does worst here
-            vfloat64m1_t result0 = __riscv_vget_v_f64m2_f64m1(result01, 0);
-            vfloat64m1_t result1 = __riscv_vget_v_f64m2_f64m1(result01, 1);
-            vfloat64m1_t result2 = __riscv_vget_v_f64m2_f64m1(result23, 0);
-            vfloat64m1_t result3 = __riscv_vget_v_f64m2_f64m1(result23, 1);
-            vfloat64m1_t result4 = __riscv_vget_v_f64m2_f64m1(result45, 0);
-            vfloat64m1_t result5 = __riscv_vget_v_f64m2_f64m1(result45, 1);
-            vfloat64m1_t result6 = __riscv_vget_v_f64m2_f64m1(result67, 0);
-            vfloat64m1_t result7 = __riscv_vget_v_f64m2_f64m1(result67, 1);
-            vfloat64m1_t result8 = __riscv_vget_v_f64m2_f64m1(result89, 0);
-            vfloat64m1_t result9 = __riscv_vget_v_f64m2_f64m1(result89, 1);
-            vfloat64m1_t result10 = __riscv_vget_v_f64m2_f64m1(resultAB, 0);
-            vfloat64m1_t result11 = __riscv_vget_v_f64m2_f64m1(resultAB, 1);
-            vfloat64m1_t result12 = __riscv_vget_v_f64m2_f64m1(resultCD, 0);
-            vfloat64m1_t result13 = __riscv_vget_v_f64m2_f64m1(resultCD, 1);
-            vfloat64m1_t result14 = __riscv_vget_v_f64m2_f64m1(resultEF, 0);
-            vfloat64m1_t result15 = __riscv_vget_v_f64m2_f64m1(resultEF, 1);
-
             FLOAT *C2 = C;
 
             vfloat64m2_t c01 = __riscv_vle64_v_f64m2(C, 8); C += ldc;

kernel/riscv64/sgemm_kernel_16x8_zvl256b.c

@@ -2137,7 +2137,7 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT* A, FLOAT* B, F
 {
     if (K <= 0) return 0;
     const BLASLONG m_edge = M & 15;
-    const bool S = (M == (ldc & 0xF));
+    const bool S = (ldc == m_edge);
 
     // -- MAIN PASS
 
@@ -2194,24 +2194,6 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT alpha, FLOAT* A, FLOAT* B, F
                 resultEF = __riscv_vfmacc_vf_f32m2( resultEF, B7, A00, 16 );
             }
 
-            // LMUL = 2 does worst here
-            vfloat32m1_t result0 = __riscv_vget_v_f32m2_f32m1(result01, 0);
-            vfloat32m1_t result1 = __riscv_vget_v_f32m2_f32m1(result01, 1);
-            vfloat32m1_t result2 = __riscv_vget_v_f32m2_f32m1(result23, 0);
-            vfloat32m1_t result3 = __riscv_vget_v_f32m2_f32m1(result23, 1);
-            vfloat32m1_t result4 = __riscv_vget_v_f32m2_f32m1(result45, 0);
-            vfloat32m1_t result5 = __riscv_vget_v_f32m2_f32m1(result45, 1);
-            vfloat32m1_t result6 = __riscv_vget_v_f32m2_f32m1(result67, 0);
-            vfloat32m1_t result7 = __riscv_vget_v_f32m2_f32m1(result67, 1);
-            vfloat32m1_t result8 = __riscv_vget_v_f32m2_f32m1(result89, 0);
-            vfloat32m1_t result9 = __riscv_vget_v_f32m2_f32m1(result89, 1);
-            vfloat32m1_t result10 = __riscv_vget_v_f32m2_f32m1(resultAB, 0);
-            vfloat32m1_t result11 = __riscv_vget_v_f32m2_f32m1(resultAB, 1);
-            vfloat32m1_t result12 = __riscv_vget_v_f32m2_f32m1(resultCD, 0);
-            vfloat32m1_t result13 = __riscv_vget_v_f32m2_f32m1(resultCD, 1);
-            vfloat32m1_t result14 = __riscv_vget_v_f32m2_f32m1(resultEF, 0);
-            vfloat32m1_t result15 = __riscv_vget_v_f32m2_f32m1(resultEF, 1);
-
             FLOAT *C2 = C;
 
             vfloat32m2_t c01 = __riscv_vle32_v_f32m2(C, 16); C += ldc;