diff --git a/hotspot/src/cpu/aarch64/vm/c1_LIRGenerator_aarch64.cpp b/hotspot/src/cpu/aarch64/vm/c1_LIRGenerator_aarch64.cpp
index 7080ea10d..62a8ab7bd 100644
--- a/hotspot/src/cpu/aarch64/vm/c1_LIRGenerator_aarch64.cpp
+++ b/hotspot/src/cpu/aarch64/vm/c1_LIRGenerator_aarch64.cpp
@@ -919,6 +919,126 @@ void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
   }
 }
 
+void LIRGenerator::do_dgemm_dgemm(Intrinsic* x) {
+  assert(x->number_of_arguments() == 16, "wrong type");
+
+  LIRItem ta(x->argument_at(0), this);
+  LIRItem tb(x->argument_at(1), this);
+  LIRItem m(x->argument_at(2), this);
+  LIRItem n(x->argument_at(3), this);
+  LIRItem k(x->argument_at(4), this);
+  LIRItem alpha(x->argument_at(5), this);
+  LIRItem a(x->argument_at(6), this);
+  LIRItem a_offset(x->argument_at(7), this);
+  LIRItem lda(x->argument_at(8), this);
+  LIRItem b(x->argument_at(9), this);
+  LIRItem b_offset(x->argument_at(10), this);
+  LIRItem ldb(x->argument_at(11), this);
+  LIRItem beta(x->argument_at(12), this);
+  LIRItem c(x->argument_at(13), this);
+  LIRItem c_offset(x->argument_at(14), this);
+  LIRItem ldc(x->argument_at(15), this);
+
+  ta.load_item();
+  tb.load_item();
+  m.load_item();
+  n.load_item();
+  k.load_item();
+  alpha.load_item();
+  a.load_item();
+  a_offset.load_nonconstant();
+  lda.load_item();
+  b.load_item();
+  b_offset.load_nonconstant();
+  ldb.load_item();
+  beta.load_item();
+  c.load_item();
+  c_offset.load_nonconstant();
+  ldc.load_item();
+
+  LIR_Opr ta_base = ta.result();
+  LIR_Opr tb_base = tb.result();
+  LIR_Opr r_m = m.result();
+  LIR_Opr r_n = n.result();
+  LIR_Opr r_k = k.result();
+  LIR_Opr r_alpha = alpha.result();
+  LIR_Opr a_base = a.result();
+  LIR_Opr r_a_offset = a_offset.result();
+  LIR_Opr r_lda = lda.result();
+  LIR_Opr b_base = b.result();
+  LIR_Opr r_b_offset = b_offset.result();
+  LIR_Opr r_ldb = ldb.result();
+  LIR_Opr r_beta = beta.result();
+  LIR_Opr c_base = c.result();
+  LIR_Opr r_c_offset = c_offset.result();
+  LIR_Opr r_ldc = ldc.result();
+
+  LIR_Opr ta_value = load_String_value(ta_base);
+  LIR_Opr ta_offset = load_String_offset(ta_base);
+  LIR_Opr tb_value = load_String_value(tb_base);
+  LIR_Opr tb_offset = load_String_offset(tb_base);
+
+  LIR_Address* addr_ta = emit_array_address(ta_value, ta_offset, T_CHAR, false);
+  LIR_Address* addr_tb = emit_array_address(tb_value, tb_offset, T_CHAR, false);
+  LIR_Address* addr_a = emit_array_address(a_base, r_a_offset, T_DOUBLE, false);
+  LIR_Address* addr_b = emit_array_address(b_base, r_b_offset, T_DOUBLE, false);
+  LIR_Address* addr_c = emit_array_address(c_base, r_c_offset, T_DOUBLE, false);
+
+  LIR_Opr tmp = new_pointer_register();
+  LIR_Opr ta_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_ta), tmp);
+  __ move(tmp, ta_addr);
+  tmp = new_pointer_register();
+  LIR_Opr tb_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_tb), tmp);
+  __ move(tmp, tb_addr);
+  tmp = new_pointer_register();
+  LIR_Opr a_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_a), tmp);
+  __ move(tmp, a_addr);
+  tmp = new_pointer_register();
+  LIR_Opr b_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_b), tmp);
+  __ move(tmp, b_addr);
+  tmp = new_pointer_register();
+  LIR_Opr c_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_c), tmp);
+  __ move(tmp, c_addr);
+
+  BasicTypeList signature(13);
+  signature.append(T_ADDRESS);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_INT);
+  signature.append(T_INT);
+  signature.append(T_DOUBLE);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_DOUBLE);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+
+  LIR_OprList* args = new LIR_OprList();
+  args->append(ta_addr);
+  args->append(tb_addr);
+  args->append(r_m);
+  args->append(r_n);
+  args->append(r_k);
+  args->append(r_alpha);
+  args->append(a_addr);
+  args->append(r_lda);
+  args->append(b_addr);
+  args->append(r_ldb);
+  args->append(r_beta);
+  args->append(c_addr);
+  args->append(r_ldc);
+
+  assert(StubRoutines::dgemmDgemm() != NULL, "invalid stub entry");
+  call_runtime(&signature, args, StubRoutines::dgemmDgemm(), voidType, NULL);
+  set_no_result(x);
+}
 
 void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
   assert(x->number_of_arguments() == 5, "wrong type");
@@ -1038,6 +1158,114 @@ void LIRGenerator::do_update_CRC32(Intrinsic* x) {
   }
 }
 
+void LIRGenerator::do_dgemv_dgemv(Intrinsic* x) {
+  assert(x->number_of_arguments() == 14, "wrong type");
+
+  LIRItem trans(x->argument_at(0), this);
+  LIRItem m(x->argument_at(1), this);
+  LIRItem n(x->argument_at(2), this);
+  LIRItem alpha(x->argument_at(3), this);
+  LIRItem array_a(x->argument_at(4), this);
+  LIRItem array_a_offset(x->argument_at(5), this);
+  LIRItem lda(x->argument_at(6), this);
+  LIRItem array_x(x->argument_at(7), this);
+  LIRItem array_x_offset(x->argument_at(8), this);
+  LIRItem incx(x->argument_at(9), this);
+  LIRItem beta(x->argument_at(10), this);
+  LIRItem array_y(x->argument_at(11), this);
+  LIRItem array_y_offset(x->argument_at(12), this);
+  LIRItem incy(x->argument_at(13), this);
+
+  trans.load_item();
+  m.load_item();
+  n.load_item();
+  alpha.load_item();
+  array_a.load_item();
+  array_a_offset.load_nonconstant();
+  lda.load_item();
+  array_x.load_item();
+  array_x_offset.load_nonconstant();
+  incx.load_item();
+  beta.load_item();
+  array_y.load_item();
+  array_y_offset.load_nonconstant();
+  incy.load_item();
+
+  LIR_Opr res_trans_base = trans.result();
+  LIR_Opr res_m = m.result();
+  LIR_Opr res_n = n.result();
+  LIR_Opr res_alpha = alpha.result();
+  LIR_Opr res_a_base = array_a.result();
+  LIR_Opr res_a_offset = array_a_offset.result();
+  LIR_Opr res_lda = lda.result();
+  LIR_Opr res_x_base = array_x.result();
+  LIR_Opr res_x_offset = array_x_offset.result();
+  LIR_Opr res_incx = incx.result();
+  LIR_Opr res_beta = beta.result();
+  LIR_Opr res_y_base = array_y.result();
+  LIR_Opr res_y_offset = array_y_offset.result();
+  LIR_Opr res_incy = incy.result();
+
+  LIR_Opr addr_trans_base = LIRGenerator::load_String_value(res_trans_base);
+  LIR_Opr addr_trans_offset = LIRGenerator::load_String_offset(res_trans_base);
+  LIR_Address* addr_trans = emit_array_address(addr_trans_base, addr_trans_offset, T_CHAR, false);
+
+  LIR_Address* addr_a = emit_array_address(res_a_base, res_a_offset, T_DOUBLE, false);
+  LIR_Address* addr_x = emit_array_address(res_x_base, res_x_offset, T_DOUBLE, false);
+  LIR_Address* addr_y = emit_array_address(res_y_base, res_y_offset, T_DOUBLE, false);
+
+  // load addr to register
+  LIR_Opr tmp = new_pointer_register();
+  LIR_Opr trans_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_trans), tmp);
+  __ move(tmp, trans_addr);
+
+  LIR_Opr tmp1 = new_pointer_register();
+  LIR_Opr a_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_a), tmp1);
+  __ move(tmp1, a_addr);
+
+  LIR_Opr tmp2 = new_pointer_register();
+  LIR_Opr x_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_x), tmp2);
+  __ move(tmp2, x_addr);
+
+  LIR_Opr tmp3 = new_pointer_register();
+  LIR_Opr y_addr = new_register(T_ADDRESS);
+  __ leal(LIR_OprFact::address(addr_y), tmp3);
+  __ move(tmp3, y_addr);
+
+  BasicTypeList signature(11);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_INT);
+  signature.append(T_DOUBLE);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+  signature.append(T_DOUBLE);
+  signature.append(T_ADDRESS);
+  signature.append(T_INT);
+
+  LIR_OprList* args = new LIR_OprList();
+  args->append(trans_addr);
+  args->append(res_m);
+  args->append(res_n);
+  args->append(res_alpha);
+  args->append(a_addr);
+  args->append(res_lda);
+  args->append(x_addr);
+  args->append(res_incx);
+  args->append(res_beta);
+  args->append(y_addr);
+  args->append(res_incy);
+
+  assert(StubRoutines::dgemvDgemv() != NULL, "invalid stub entry");
+  call_runtime(&signature, args, StubRoutines::dgemvDgemv(), voidType, NULL);
+  set_no_result(x);
+}
+
 // _i2l, _i2f, _i2d, _l2i, _l2f, _l2d, _f2i, _f2l, _f2d, _d2i, _d2l, _d2f
 // _i2b, _i2c, _i2s
 void LIRGenerator::do_Convert(Convert* x) {
diff --git a/hotspot/src/cpu/aarch64/vm/interpreterGenerator_aarch64.hpp b/hotspot/src/cpu/aarch64/vm/interpreterGenerator_aarch64.hpp
index c0aaa1de4..a275a6a99 100644
--- a/hotspot/src/cpu/aarch64/vm/interpreterGenerator_aarch64.hpp
+++ b/hotspot/src/cpu/aarch64/vm/interpreterGenerator_aarch64.hpp
@@ -50,6 +50,11 @@ void generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpa
   address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind);
   void lock_method(void);
   void generate_stack_overflow_check(void);
+  void load_String_value(Register src, Register dst);
+  void load_String_offset(Register src, Register dst);
+  void emit_array_address(Register src, Register idx, Register dst, BasicType type);
+  address generate_Dgemm_dgemm_entry();
+  address generate_Dgemv_dgemv_entry();
 
   void generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue);
   void generate_counter_overflow(Label* do_continue);
diff --git a/hotspot/src/cpu/aarch64/vm/stubGenerator_aarch64.cpp b/hotspot/src/cpu/aarch64/vm/stubGenerator_aarch64.cpp
index c5ec637a1..125983179 100644
--- a/hotspot/src/cpu/aarch64/vm/stubGenerator_aarch64.cpp
+++ b/hotspot/src/cpu/aarch64/vm/stubGenerator_aarch64.cpp
@@ -3221,6 +3221,44 @@ class StubGenerator: public StubCodeGenerator {
     return start;
   }
 
+  address load_BLAS_library() {
+    // Try to load BLAS library.
+    const char library_name[] = "openblas";
+    char err_buf[1024] = {0};
+    char path[JVM_MAXPATHLEN] = {0};
+    os::jvm_path(path, sizeof(path));
+    int jvm_offset = -1;
+
+    // Match "jvm[^/]*" in jvm_path.
+    const char* last_name = strrchr(path, '/');
+    last_name = last_name ? last_name : path;
+    const char* last_lib_name = strstr(last_name, "jvm");
+    if (last_lib_name != NULL) {
+      jvm_offset = last_lib_name - path;
+    }
+
+    address library = NULL;
+    // Find the BLAS shared library.
+    // Search path: <home>/jre/lib/<arch>/<vm>/libopenblas.so
+    if (jvm_offset >= 0) {
+      if (jvm_offset + strlen(library_name) + strlen(os::dll_file_extension()) < JVM_MAXPATHLEN) {
+        strncpy(&path[jvm_offset], library_name, JVM_MAXPATHLEN - jvm_offset);
+        strncat(path, os::dll_file_extension(), strlen(os::dll_file_extension()));
+        library = (address)os::dll_load(path, err_buf, sizeof(err_buf));
+      }
+    }
+    return library;
+  }
+
+  address get_BLAS_func_entry(address library, const char* func_name) {
+    if (library == NULL) {
+        return NULL;
+    }
+
+    // Try to find BLAS function entry.
+    return (address)os::dll_lookup((void*)library, func_name);
+  }
+
   /**
    *  Arguments:
    *
@@ -3254,6 +3292,218 @@ class StubGenerator: public StubCodeGenerator {
     return start;
   }
 
+  // Parameter conversion from JVM to native BLAS
+  //
+  // Register:
+  // r0: transa                         r0: transa
+  // r1: transb                         r1: transb
+  // r2: m                              r2: &m
+  // r3: n                              r3: &n
+  // r4: k            =========>        r4: &k
+  // r5: A                              r5: &alpha
+  // r6: lda                            r6: A
+  // r7: B                              r7: &lda
+  // v0: alpha
+  // v1: beta
+  //
+  // Stack:
+  // |-------|                          |-------|
+  // | ldc   |                          | ldc   |
+  // |-------|                          |-------|
+  // | C     |                          | C     |
+  // |-------|                          |-------|
+  // | ldb   |                          | ldb   |
+  // |-------| <-- sp                   |-------|
+  // |       |                          | m     |
+  // |-------|                          |-------|
+  // |       |                          | n     |
+  // |-------|                          |-------|
+  // |       |                          | k     |
+  // |-------|                          |-------|
+  // |       |                          | lda   |
+  // |-------|                          |-------|
+  // |       |                          | alpha |
+  // |-------|                          |-------|
+  // |       |                          | beta  |
+  // |-------|        =========>        |-------|
+  // |       |                          | lr    |
+  // |-------|                          |-------|
+  // |       |                          | rfp   |
+  // |-------|                          |-------| <-- fp
+  // | ...   |                          | ...   |
+  // |-------|                          |-------|
+  // |       |                          | &ldc  |
+  // |-------|                          |-------|
+  // |       |                          | C     |
+  // |-------|                          |-------|
+  // |       |                          | &bata |
+  // |-------|                          |-------|
+  // |       |                          | &ldb  |
+  // |-------|                          |-------|
+  // |       |                          | B     |
+  // |-------|                          |-------| <-- sp
+  address generate_dgemmDgemm(address library) {
+    __ align(CodeEntryAlignment);
+    StubCodeMark mark(this, "StubRoutines", "dgemm_dgemm");
+
+    address fn = get_BLAS_func_entry(library, "dgemm_");
+    if (fn == NULL) return NULL;
+
+    address start = __ pc();
+
+    const Register transa       = c_rarg0;
+    const Register transb       = c_rarg1;
+    const Register m            = c_rarg2;
+    const Register n            = c_rarg3;
+    const Register k            = c_rarg4;
+    const FloatRegister alpha   = c_farg0;
+    const Register A            = c_rarg5;
+    const Register lda          = c_rarg6;
+    const Register B            = c_rarg7;
+    const FloatRegister beta    = c_farg1;
+
+    BLOCK_COMMENT("Entry:");
+
+    // extend stack
+    __ sub(sp, sp, 0x60);
+    __ stp(rfp, lr, Address(sp, 48));
+    __ add(rfp, sp, 0x30);
+    // load BLAS function entry
+    __ mov(rscratch1, fn);
+    // C
+    __ ldr(rscratch2, Address(rfp, 56));
+    // store m / n to stack
+    __ stpw(n, m, Address(rfp, 40));
+    // &beta
+    __ add(r2, rfp, 0x10);
+    // store k / lda to stack
+    __ stpw(lda, k, Address(rfp, 32));
+    // load ldc
+    __ add(r3, rfp, 0x40);
+    // store C / &beta
+    __ stp(r2, rscratch2, Address(sp, 16));
+    // &ldb
+    __ add(r2, rfp, 0x30);
+    // store B
+    __ str(B, Address(sp));
+    // move A from r5 to r6
+    __ mov(r6, A);
+    // store ldc
+    __ str(r3, Address(sp, 32));
+    // &alpha
+    __ add(r5, rfp, 0x18);
+    // store &ldb
+    __ str(r2, Address(sp, 8));
+    // &k
+    __ add(r4, rfp, 0x24);
+    // store alpha / beta
+    __ stpd(beta, alpha, Address(rfp, 16));
+    // load &lda to r7
+    __ add(r7, rfp, 0x20);
+    // load &n
+    __ add(r3, rfp, 0x28);
+    // load &m
+    __ add(r2, rfp, 0x2c);
+    // call dgemm
+    __ blr(rscratch1);
+
+    // restore rfp and lr
+    __ ldp(rfp, lr, Address(sp, 48));
+    // exit stack
+    __ add(sp, sp, 0x60);
+    __ ret(lr);
+
+    return start;
+  }
+
+  /**
+   *  public void dgemv(String trans, int m, int n,
+   *                    double alpha, double[] a, int lda,
+   *                    double[] x, int incx,
+   *                    double beta, double[] y, int incy)
+   *
+   *  Arguments:
+   *
+   *  Inputs:
+   *   c_rarg0         - char* trans
+   *   c_rarg1         - int m
+   *   c_rarg2         - int n
+   *   d0/c_farg0      - double alpha
+   *   c_rarg3         - double[] a
+   *   c_rarg4         - int lda
+   *   c_rarg5         - double[] x
+   *   c_rarg6         - int incx
+   *   d1/c_farg1      - double beta
+   *   c_rarg7         - double[] y
+   *   [sp]            - int incy
+   *
+   *  Output:
+   *       null
+   *
+   */
+
+  address generate_dgemvDgemv(address library) {
+    __ align(CodeEntryAlignment);
+    StubCodeMark mark(this, "StubRoutines", "dgemv_dgemv");
+
+    address fn = get_BLAS_func_entry(library, "dgemv_");
+    if (fn == NULL) return NULL;
+
+    address start = __ pc();
+    BLOCK_COMMENT("Entry: ");
+
+    Register trans = c_rarg0;
+    Register m = c_rarg1;
+    Register n = c_rarg2;
+    Register a = c_rarg3;
+    Register lda = c_rarg4;
+    Register x = c_rarg5;
+    Register incx = c_rarg6;
+    Register y = c_rarg7;
+
+    FloatRegister alpha = c_farg0;
+    FloatRegister beta = c_farg1;
+
+    __ sub(sp, sp, 0x50);
+    __ stp(rfp, lr, Address(sp, 32));
+    __ add(rfp, sp, 0x20);
+
+    // no need for saving trans to tmp register, keep it in register x0
+    __ strw(m, Address(rfp, 44));
+    __ strw(n, Address(rfp, 40));
+    __ strd(alpha, Address(rfp, 32));
+    __ strw(lda, Address(rfp, 28));
+    __ strw(incx, Address(rfp, 24));
+    __ strd(beta, Address(rfp, 16));
+
+    // pre call
+    // load incy and push on stack, order incy --> y --> beta
+    __ add(r1, rfp, 0x30);
+    __ str(r1, Address(sp, 16));
+    __ str(y, Address(sp, 8));
+    __ add(r1, rfp, 0x10);
+    __ str(r1, Address(sp));
+
+    __ add(r7, rfp, 0x18);
+    __ mov(r6, x);
+    __ add(r5, rfp, 0x1c);
+    __ mov(r4, a);
+    __ add(r3, rfp, 0x20);
+    __ add(r2, rfp, 0x28);
+    __ add(r1, rfp, 0x2c);
+
+    __ mov(rscratch1, fn);
+    __ blr(rscratch1);
+
+    __ ldp(rfp, lr, Address(sp, 32));
+    __ add(sp, sp, 0x50);
+    __ ret(lr);
+
+    return start;
+  }
+
+
+
   /**
    *  Arguments:
    *
@@ -4252,6 +4502,14 @@ class StubGenerator: public StubCodeGenerator {
       StubRoutines::_crc_table_adr = (address)StubRoutines::aarch64::_crc_table;
       StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
     }
+
+    if (UseF2jBLASIntrinsics) {
+      StubRoutines::_BLAS_library = load_BLAS_library();
+      // F2jBLAS intrinsics will use the implements in BLAS dynamic library
+      StubRoutines::_ddotF2jBLAS = generate_ddotF2jBLAS();
+      StubRoutines::_dgemmDgemm = generate_dgemmDgemm(StubRoutines::_BLAS_library);
+      StubRoutines::_dgemvDgemv = generate_dgemvDgemv(StubRoutines::_BLAS_library);
+    }
   }
 
   void generate_all() {
@@ -4296,10 +4554,6 @@ class StubGenerator: public StubCodeGenerator {
       StubRoutines::_montgomerySquare = g.generate_multiply();
     }
 
-    if (UseF2jBLASIntrinsics) {
-      StubRoutines::_ddotF2jBLAS = generate_ddotF2jBLAS();
-    }
-
     if (UseAESIntrinsics) {
       StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
       StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
diff --git a/hotspot/src/cpu/aarch64/vm/templateInterpreter_aarch64.cpp b/hotspot/src/cpu/aarch64/vm/templateInterpreter_aarch64.cpp
index ae5cb3f32..924b6670f 100644
--- a/hotspot/src/cpu/aarch64/vm/templateInterpreter_aarch64.cpp
+++ b/hotspot/src/cpu/aarch64/vm/templateInterpreter_aarch64.cpp
@@ -856,6 +856,250 @@ address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpret
   return generate_native_entry(false);
 }
 
+// Access the char-array of String
+void InterpreterGenerator::load_String_value(Register src, Register dst) {
+  //  Need to cooperate with JDK-8243996
+  int value_offset = java_lang_String::value_offset_in_bytes();
+
+  __ add(src, src, value_offset);
+  __ load_heap_oop(dst, Address(src));
+}
+
+void InterpreterGenerator::load_String_offset(Register src, Register dst) {
+  __ mov(dst, 0);
+
+  // Get String value offset, because of order of initialization for Interpreter,
+  // we have to hardcode the offset for String value. (JDK-8243996)
+  if (java_lang_String::has_offset_field()) {
+    int offset_offset = java_lang_String::offset_offset_in_bytes();
+    __ add(src, src, offset_offset);
+    __ ldrw(dst, Address(src));
+  }
+}
+
+void InterpreterGenerator::emit_array_address(Register src, Register idx,
+                                              Register dst, BasicType type) {
+  int offset_in_bytes = arrayOopDesc::base_offset_in_bytes(type);
+  int elem_size = type2aelembytes(type);
+  int shift = exact_log2(elem_size);
+
+  __ lsl(idx, idx, shift);
+  __ add(idx, idx, offset_in_bytes);
+  __ add(dst, src, idx);
+}
+
+/**
+ * Stub Arguments:
+ *
+ *   c_rarg0   - char* transa
+ *   c_rarg1   - char* transb
+ *   c_rarg2   - int m
+ *   c_rarg3   - int n
+ *   c_rarg4   - int k
+ *   d0        - double alpha
+ *   c_rarg5   - double[] A
+ *   c_rarg6   - int lda
+ *   c_rarg7   - double[] B
+ *   d1        - double beta
+ *   [sp + 16] - int ldc
+ *   [sp + 8]  - double[] C
+ *   [sp]      - int ldb
+ *
+ */
+address InterpreterGenerator::generate_Dgemm_dgemm_entry() {
+  if (!UseF2jBLASIntrinsics || (StubRoutines::dgemmDgemm() == NULL)) return NULL;
+
+  address entry = __ pc();
+
+  // r13: senderSP must preserved for slow path
+
+  // Arguments are reversed on java expression stack
+  const Register ta         = c_rarg0;
+  const Register tb         = c_rarg1;
+  const Register m          = c_rarg2;
+  const Register n          = c_rarg3;
+  const Register k          = c_rarg4;
+  const FloatRegister alpha = c_farg0;
+  const Register A          = c_rarg5;
+  const Register lda        = c_rarg6;
+  const Register B          = c_rarg7;
+  const FloatRegister beta  = c_farg1;
+  const Register tmp1       = rscratch1;
+  const Register tmp2       = rscratch2;
+
+  // trana
+  __ ldr(ta, Address(esp, 17 * wordSize));
+  load_String_value(ta, tmp1);
+  load_String_offset(ta, tmp2);
+  emit_array_address(tmp1, tmp2, ta, T_CHAR);
+  // tranb
+  __ ldr(tb, Address(esp, 16 * wordSize));
+  load_String_value(tb, tmp1);
+  load_String_offset(tb, tmp2);
+  emit_array_address(tmp1, tmp2, tb, T_CHAR);
+  // m, n, k
+  __ ldrw(m, Address(esp, 15 * wordSize));
+  __ ldrw(n, Address(esp, 14 * wordSize));
+  __ ldrw(k, Address(esp, 13 * wordSize));
+  // alpha
+  __ ldrd(alpha, Address(esp, 11 * wordSize));
+  // A
+  __ ldr(tmp1, Address(esp, 10 * wordSize));
+  __ mov(tmp2, 0);
+  __ ldrw(tmp2, Address(esp, 9 * wordSize));
+  emit_array_address(tmp1, tmp2, A, T_DOUBLE);
+  // lda
+  __ ldrw(lda, Address(esp, 8 * wordSize));
+  // B
+  __ ldr(tmp1, Address(esp, 7 * wordSize));
+  __ ldrw(tmp2, Address(esp, 6 * wordSize));
+  emit_array_address(tmp1, tmp2, B, T_DOUBLE);
+  // beta
+  __ ldrd(beta, Address(esp, 3 * wordSize));
+  // Start pushing arguments to machine stack.
+  //
+  // Remove the incoming args, peeling the machine SP back to where it
+  // was in the caller.  This is not strictly necessary, but unless we
+  // do so the stack frame may have a garbage FP; this ensures a
+  // correct call stack that we can always unwind.  The ANDR should be
+  // unnecessary because the sender SP in r13 is always aligned, but
+  // it doesn't hurt.
+  __ andr(sp, r13, -16);
+  __ str(lr, Address(sp, -wordSize));
+  // ldc
+  __ ldrw(tmp1, Address(esp, 0x0));
+  __ strw(tmp1, Address(sp, 2 * -wordSize));
+  // C
+  __ ldr(tmp1, Address(esp, 2 * wordSize));
+  __ ldrw(tmp2, Address(esp, wordSize));
+  emit_array_address(tmp1, tmp2, tmp1, T_DOUBLE);
+  __ str(tmp1, Address(sp, 3 * -wordSize));
+  // ldb
+  __ ldrw(tmp2, Address(esp, 5 * wordSize));
+  __ strw(tmp2, Address(sp, 4 * -wordSize));
+
+  // Call function
+  __ add(sp, sp, 4 * -wordSize);
+  address fn = CAST_FROM_FN_PTR(address, StubRoutines::dgemmDgemm());
+  __ mov(tmp1, fn);
+  __ blr(tmp1);
+
+  __ ldr(lr, Address(sp, 3 * wordSize));
+  // For assert(Rd != sp || imm % 16 == 0)
+  __ add(sp, sp, 4 * wordSize);
+  __ br(lr);
+
+  return entry;
+}
+
+address InterpreterGenerator::generate_Dgemv_dgemv_entry() {
+  if (StubRoutines::dgemvDgemv() == NULL) return NULL;
+  address entry = __ pc();
+
+  const Register trans = c_rarg0;              // trans
+  const Register m = c_rarg1;                  // m
+  const Register n = c_rarg2;                  // n
+  const Register a = c_rarg3;                  // array a addr
+  const Register lda = c_rarg4;                // lda
+  const Register x = c_rarg5;                  // array x addr
+  const Register incx = c_rarg6;               // incx
+  const Register y = c_rarg7;                  // array y addr
+
+  const FloatRegister alpha = v0;              // alpha
+  const FloatRegister beta = v1;               // beta
+
+  const Register tmp1 = rscratch1;
+  const Register tmp2 = rscratch2;
+
+  // esp: expression stack of caller
+  // dgemv parameter ---> the position in stack ---> move to register
+  // | char* trans  |        | esp + 15 |                |  r0  |
+  // | int m        |        | esp + 14 |                |  r1  |
+  // | int n        |        | esp + 13 |                |  r2  |
+  // | double alpha |        | esp + 11 |                |  v0  |
+  // ----------------        ------------                --------
+  // | double* a    |        | esp + 10 |                |      |
+  // |              |        |          |                |  r3  |
+  // | int a_offset |        | esp + 9  |                |      |
+  // ----------------        ------------                --------
+  // | int lda      |        | esp + 8  |                |  r4  |
+  // ----------------        ------------                --------
+  // | double* x    |        | esp + 7  |                |      |
+  // |              |        |          |                |  r5  |
+  // | int x_offset |        | esp + 6  |                |      |
+  // ----------------        ------------                --------
+  // | int incx     |        | esp + 5  |                |  r6  |
+  // | double beta  |        | esp + 3  |                |  v1  |
+  // ----------------        ------------                --------
+  // | double* y    |        | esp + 2  |                |      |
+  // |              |        |          |                |  r7  |
+  // | int y_offset |        | esp + 1  |                |      |
+  // ----------------        ------------                --------
+  // | int incy     |        | esp      |                | [sp] |
+
+
+  // trans
+  __ ldr(trans, Address(esp, 15 * wordSize));
+  load_String_value(trans, tmp1);
+  load_String_offset(trans, tmp2);
+  emit_array_address(tmp1, tmp2, trans, T_CHAR);
+  // m, n
+  __ ldrw(m, Address(esp, 14 * wordSize));
+  __ ldrw(n, Address(esp, 13 * wordSize));
+
+  // alpha
+  __ ldrd(alpha, Address(esp, 11 * wordSize));
+
+  // a
+  __ ldr(tmp1, Address(esp, 10 * wordSize));
+  __ mov(tmp2, zr);
+  __ ldrw(tmp2, Address(esp, 9 * wordSize));
+  emit_array_address(tmp1, tmp2, a, T_DOUBLE);
+
+  // lda
+  __ ldrw(lda, Address(esp, 8 * wordSize));
+
+  // x
+  __ ldr(tmp1, Address(esp, 7 * wordSize));
+  __ mov(tmp2, zr);
+  __ ldrw(tmp2, Address(esp, 6 * wordSize));
+  emit_array_address(tmp1, tmp2, x, T_DOUBLE);
+
+  // incx
+  __ ldrw(incx, Address(esp, 5 * wordSize));
+
+  // beta
+  __ ldrd(beta, Address(esp, 3 * wordSize));
+
+  // y
+  __ ldr(tmp1, Address(esp, 2 * wordSize));
+  __ mov(tmp2, zr);
+  __ ldrw(tmp2, Address(esp, wordSize));
+  emit_array_address(tmp1, tmp2, y, T_DOUBLE);
+
+  // resume sp, restore lr
+  __ andr(sp, r13, -16);
+  __ str(lr, Address(sp, -wordSize));
+
+  // incy, push on stack
+  __ ldrw(tmp1, Address(esp, 0));
+  __ strw(tmp1, Address(sp, 2 * -wordSize));
+
+  __ add(sp, sp, -2 * wordSize);
+
+  // call function
+  address fn = CAST_FROM_FN_PTR(address, StubRoutines::dgemvDgemv());
+  __ mov(tmp1, fn);
+  __ blr(tmp1);
+
+  // resume lr
+  __ ldr(lr, Address(sp, wordSize));
+  __ add(sp, sp, 2 * wordSize);
+  __ br(lr);
+
+  return  entry;
+}
+
 void InterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
   // Bang each page in the shadow zone. We can't assume it's been done for
   // an interpreter frame with greater than a page of locals, so each page
@@ -1575,6 +1819,10 @@ address AbstractInterpreterGenerator::generate_method_entry(
                                            : // fall thru
   case Interpreter::java_util_zip_CRC32_updateByteBuffer
                                            : entry_point = ((InterpreterGenerator*)this)->generate_CRC32_updateBytes_entry(kind); break;
+  case Interpreter::org_netlib_blas_Dgemm_dgemm
+                                           : entry_point = ((InterpreterGenerator*)this)->generate_Dgemm_dgemm_entry(); break;
+  case Interpreter::org_netlib_blas_Dgemv_dgemv
+                                           : entry_point = ((InterpreterGenerator*)this)->generate_Dgemv_dgemv_entry(); break;
   default                                  : ShouldNotReachHere();                                                       break;
   }
 
diff --git a/hotspot/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp b/hotspot/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp
index f1160792a..477c6e550 100644
--- a/hotspot/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp
+++ b/hotspot/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp
@@ -754,6 +754,13 @@ void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
   }
 }
 
+void LIRGenerator::do_dgemm_dgemm(Intrinsic* x) {
+  fatal("BLAS intrinsics are not implemented on this platform!");
+}
+
+void LIRGenerator::do_dgemv_dgemv(Intrinsic* x) {
+  fatal("BLAS intrinsics are not implemented on this platform!");
+}
 
 void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
   assert(x->number_of_arguments() == 5, "wrong type");
diff --git a/hotspot/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp b/hotspot/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp
index dd23f005b..d1ecbaeb4 100644
--- a/hotspot/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp
+++ b/hotspot/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp
@@ -896,6 +896,13 @@ void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
   }
 }
 
+void LIRGenerator::do_dgemm_dgemm(Intrinsic* x) {
+  fatal("BLAS intrinsics are not implemented on this platform!");
+}
+
+void LIRGenerator::do_dgemv_dgemv(Intrinsic *x) {
+  fatal("Blas intrinsics are not implemented on this platform!");
+}
 
 void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
   assert(x->number_of_arguments() == 5, "wrong type");
diff --git a/hotspot/src/share/vm/c1/c1_GraphBuilder.cpp b/hotspot/src/share/vm/c1/c1_GraphBuilder.cpp
index 459315cb7..79b2b2bb1 100644
--- a/hotspot/src/share/vm/c1/c1_GraphBuilder.cpp
+++ b/hotspot/src/share/vm/c1/c1_GraphBuilder.cpp
@@ -3672,6 +3672,20 @@ bool GraphBuilder::try_inline_intrinsics(ciMethod* callee) {
     case vmIntrinsics::_fullFence :
       break;
 
+    case vmIntrinsics::_dgemm_dgemm:
+      if (!UseF2jBLASIntrinsics || (StubRoutines::dgemmDgemm() == NULL)) {
+        return false;
+      }
+      cantrap = false;
+      preserves_state = true;
+      break;
+
+    case vmIntrinsics::_dgemv_dgemv:
+      if (!UseF2jBLASIntrinsics || (StubRoutines::dgemvDgemv() == NULL)) return false;
+      cantrap = false;
+      preserves_state = true;
+      break;
+
     default                       : return false; // do not inline
   }
   // create intrinsic node
diff --git a/hotspot/src/share/vm/c1/c1_LIRGenerator.cpp b/hotspot/src/share/vm/c1/c1_LIRGenerator.cpp
index 65c04e3e5..070fd8052 100644
--- a/hotspot/src/share/vm/c1/c1_LIRGenerator.cpp
+++ b/hotspot/src/share/vm/c1/c1_LIRGenerator.cpp
@@ -1208,7 +1208,7 @@ void LIRGenerator::do_Return(Return* x) {
   set_no_result(x);
 }
 
-// Examble: ref.get()
+// Example: ref.get()
 // Combination of LoadField and g1 pre-write barrier
 void LIRGenerator::do_Reference_get(Intrinsic* x) {
 
@@ -1220,7 +1220,7 @@ void LIRGenerator::do_Reference_get(Intrinsic* x) {
   LIRItem reference(x->argument_at(0), this);
   reference.load_item();
 
-  // need to perform the null check on the reference objecy
+  // need to perform the null check on the reference object
   CodeEmitInfo* info = NULL;
   if (x->needs_null_check()) {
     info = state_for(x);
@@ -1422,6 +1422,35 @@ LIR_Opr LIRGenerator::load_constant(LIR_Const* c) {
   return result;
 }
 
+// Access the char-array of String
+LIR_Opr LIRGenerator::load_String_value(LIR_Opr str) {
+  int value_offset = java_lang_String::value_offset_in_bytes();
+  LIR_Opr value = new_register(T_ARRAY);
+  LIR_Opr tmp = new_pointer_register();
+
+  __ add(str, LIR_OprFact::intConst(value_offset), tmp);
+  LIR_Address* array_addr = new LIR_Address(tmp, T_ARRAY);
+  __ load(array_addr, value);
+
+  return value;
+}
+
+LIR_Opr LIRGenerator::load_String_offset(LIR_Opr str) {
+  LIR_Opr offset = new_register(T_INT);
+
+  if (java_lang_String::has_offset_field()) {
+    LIR_Opr tmp = new_pointer_register();
+    int offset_offset = java_lang_String::offset_offset_in_bytes();
+    __ add(str, LIR_OprFact::intConst(offset_offset), tmp);
+    LIR_Address* addr = new LIR_Address(tmp, T_INT);
+    __ load(addr, offset);
+  } else {
+    offset = LIR_OprFact::intConst(0);
+  }
+
+  return offset;
+}
+
 // Various barriers
 
 void LIRGenerator::pre_barrier(LIR_Opr addr_opr, LIR_Opr pre_val,
@@ -3290,6 +3328,14 @@ void LIRGenerator::do_Intrinsic(Intrinsic* x) {
     do_update_CRC32(x);
     break;
 
+  case vmIntrinsics::_dgemm_dgemm:
+    do_dgemm_dgemm(x);
+    break;
+
+  case vmIntrinsics::_dgemv_dgemv:
+    do_dgemv_dgemv(x);
+    break;
+
   default: ShouldNotReachHere(); break;
   }
 }
diff --git a/hotspot/src/share/vm/c1/c1_LIRGenerator.hpp b/hotspot/src/share/vm/c1/c1_LIRGenerator.hpp
index 24d072b36..57d675c5b 100644
--- a/hotspot/src/share/vm/c1/c1_LIRGenerator.hpp
+++ b/hotspot/src/share/vm/c1/c1_LIRGenerator.hpp
@@ -210,6 +210,10 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   // Given an immediate value, return an operand usable in logical ops.
   LIR_Opr load_immediate(int x, BasicType type);
 
+  // Get String value and offset
+  LIR_Opr load_String_value(LIR_Opr str);
+  LIR_Opr load_String_offset(LIR_Opr str);
+
   void  set_result(Value x, LIR_Opr opr)           {
     assert(opr->is_valid(), "must set to valid value");
     assert(x->operand()->is_illegal(), "operand should never change");
@@ -251,6 +255,8 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   void do_FPIntrinsics(Intrinsic* x);
   void do_Reference_get(Intrinsic* x);
   void do_update_CRC32(Intrinsic* x);
+  void do_dgemm_dgemm(Intrinsic* x);
+  void do_dgemv_dgemv(Intrinsic* x);
 
   void do_UnsafePrefetch(UnsafePrefetch* x, bool is_store);
 
diff --git a/hotspot/src/share/vm/c1/c1_Runtime1.cpp b/hotspot/src/share/vm/c1/c1_Runtime1.cpp
index f379a0395..3ece7f6ea 100644
--- a/hotspot/src/share/vm/c1/c1_Runtime1.cpp
+++ b/hotspot/src/share/vm/c1/c1_Runtime1.cpp
@@ -305,6 +305,8 @@ const char* Runtime1::name_for_address(address entry) {
   FUNCTION_CASE(entry, JFR_TIME_FUNCTION);
 #endif
   FUNCTION_CASE(entry, StubRoutines::updateBytesCRC32());
+  FUNCTION_CASE(entry, StubRoutines::dgemmDgemm());
+  FUNCTION_CASE(entry, StubRoutines::dgemvDgemv());
 
 #undef FUNCTION_CASE
 
diff --git a/hotspot/src/share/vm/classfile/vmSymbols.cpp b/hotspot/src/share/vm/classfile/vmSymbols.cpp
index a5f89dbf8..34514022a 100644
--- a/hotspot/src/share/vm/classfile/vmSymbols.cpp
+++ b/hotspot/src/share/vm/classfile/vmSymbols.cpp
@@ -333,6 +333,8 @@ bool vmIntrinsics::should_be_pinned(vmIntrinsics::ID id) {
 #endif
   case vmIntrinsics::_currentTimeMillis:
   case vmIntrinsics::_nanoTime:
+  case vmIntrinsics::_dgemm_dgemm:
+  case vmIntrinsics::_dgemv_dgemv:
     return true;
   default:
     return false;
diff --git a/hotspot/src/share/vm/classfile/vmSymbols.hpp b/hotspot/src/share/vm/classfile/vmSymbols.hpp
index 6bd8dbedd..942d172a1 100644
--- a/hotspot/src/share/vm/classfile/vmSymbols.hpp
+++ b/hotspot/src/share/vm/classfile/vmSymbols.hpp
@@ -857,6 +857,14 @@
   do_intrinsic(_f2jblas_ddot, com_github_fommil_netlib_f2jblas, ddot_name, ddot_signature, F_R)                         \
    do_name(     ddot_name,                                         "ddot")                                              \
    do_signature(ddot_signature,                                    "(I[DI[DI)D")                                        \
+  do_class(org_netlib_blas_dgemm,                       "org/netlib/blas/Dgemm")                                        \
+  do_intrinsic(_dgemm_dgemm, org_netlib_blas_dgemm, dgemm_name, dgemm_signature, F_S)                                    \
+   do_name(     dgemm_name,                                        "dgemm")                                             \
+   do_signature(dgemm_signature,                                   "(Ljava/lang/String;Ljava/lang/String;IIID[DII[DIID[DII)V")  \
+  do_class(org_netlib_blas_dgemv,                       "org/netlib/blas/Dgemv")                                        \
+  do_intrinsic(_dgemv_dgemv, org_netlib_blas_dgemv, dgemv_name, dgemv_signature, F_S)                                   \
+   do_name(     dgemv_name,                                         "dgemv")                                            \
+   do_signature(dgemv_signature,                                    "(Ljava/lang/String;IID[DII[DIID[DII)V")            \
                                                                                                                         \
   /* support for sun.security.provider.SHA2 */                                                                          \
   do_class(sun_security_provider_sha2,                             "sun/security/provider/SHA2")                        \
diff --git a/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp b/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp
index e14c50bf0..293382b3c 100644
--- a/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp
+++ b/hotspot/src/share/vm/interpreter/abstractInterpreter.hpp
@@ -100,6 +100,8 @@ class AbstractInterpreter: AllStatic {
     java_util_zip_CRC32_update,                                 // implementation of java.util.zip.CRC32.update()
     java_util_zip_CRC32_updateBytes,                            // implementation of java.util.zip.CRC32.updateBytes()
     java_util_zip_CRC32_updateByteBuffer,                       // implementation of java.util.zip.CRC32.updateByteBuffer()
+    org_netlib_blas_Dgemm_dgemm,                                // implementation of org.netlib.blas.Dgemm.dgemm()
+    org_netlib_blas_Dgemv_dgemv,                                // implementation of org.netlib.blas.Dgemv.dgemv()
     number_of_method_entries,
     invalid = -1
   };
diff --git a/hotspot/src/share/vm/interpreter/cppInterpreter.cpp b/hotspot/src/share/vm/interpreter/cppInterpreter.cpp
index 0007aa8be..9e48a1d94 100644
--- a/hotspot/src/share/vm/interpreter/cppInterpreter.cpp
+++ b/hotspot/src/share/vm/interpreter/cppInterpreter.cpp
@@ -31,17 +31,20 @@
 #ifdef CC_INTERP
 # define __ _masm->
 
-void CppInterpreter::initialize() {
+void CppInterpreter::initialize_stub() {
   if (_code != NULL) return;
+  int code_size = InterpreterCodeSize;
+  NOT_PRODUCT(code_size *= 4;)  // debug uses extra interpreter code space
+  _code = new StubQueue(new InterpreterCodeletInterface, code_size, NULL,
+                        "Interpreter");
+}
+
+void CppInterpreter::initialize_code() {
   AbstractInterpreter::initialize();
 
   // generate interpreter
   { ResourceMark rm;
     TraceTime timer("Interpreter generation", TraceStartupTime);
-    int code_size = InterpreterCodeSize;
-    NOT_PRODUCT(code_size *= 4;)  // debug uses extra interpreter code space
-    _code = new StubQueue(new InterpreterCodeletInterface, code_size, NULL,
-                          "Interpreter");
     InterpreterGenerator g(_code);
     if (PrintInterpreter) print();
   }
diff --git a/hotspot/src/share/vm/interpreter/cppInterpreter.hpp b/hotspot/src/share/vm/interpreter/cppInterpreter.hpp
index 6a6447503..58efcfaf2 100644
--- a/hotspot/src/share/vm/interpreter/cppInterpreter.hpp
+++ b/hotspot/src/share/vm/interpreter/cppInterpreter.hpp
@@ -54,7 +54,8 @@ class CppInterpreter: public AbstractInterpreter {
 
  public:
   // Initialization/debugging
-  static void       initialize();
+  static void       initialize_stub();
+  static void       initialize_code();
   // this only returns whether a pc is within generated code for the interpreter.
 
   // This is a moderately dubious interface for the c++ interpreter. Only
diff --git a/hotspot/src/share/vm/interpreter/interpreter.cpp b/hotspot/src/share/vm/interpreter/interpreter.cpp
index 7ce4bdbb3..a313f2e63 100644
--- a/hotspot/src/share/vm/interpreter/interpreter.cpp
+++ b/hotspot/src/share/vm/interpreter/interpreter.cpp
@@ -85,8 +85,6 @@ void InterpreterCodelet::print_on(outputStream* st) const {
 // Implementation of  platform independent aspects of Interpreter
 
 void AbstractInterpreter::initialize() {
-  if (_code != NULL) return;
-
   // make sure 'imported' classes are initialized
   if (CountBytecodes || TraceBytecodes || StopInterpreterAt) BytecodeCounter::reset();
   if (PrintBytecodeHistogram)                                BytecodeHistogram::reset();
@@ -114,8 +112,22 @@ void AbstractInterpreter::print() {
 }
 
 
-void interpreter_init() {
-  Interpreter::initialize();
+// The reason that interpreter initialization is split into two parts is that the first part
+// needs to run before methods are loaded (which with CDS implies linked also), and the other
+// part needs to run after. The reason is that when methods are loaded (with CDS) or linked
+// (without CDS), the i2c adapters are generated that assert we are currently in the interpreter.
+// Asserting that requires knowledge about where the interpreter is in memory. Therefore,
+// establishing the interpreter address must be done before methods are loaded. However,
+// we would like to actually generate the interpreter after methods are loaded. That allows
+// us to remove otherwise hardcoded offsets regarding fields that are needed in the interpreter
+// code. This leads to a split if 1. reserving the memory for the interpreter, 2. loading methods
+// and 3. generating the interpreter.
+void interpreter_init_stub() {
+  Interpreter::initialize_stub();
+}
+
+void interpreter_init_code() {
+  Interpreter::initialize_code();
 #ifndef PRODUCT
   if (TraceBytecodes) BytecodeTracer::set_closure(BytecodeTracer::std_closure());
 #endif // PRODUCT
@@ -251,6 +263,13 @@ AbstractInterpreter::MethodKind AbstractInterpreter::method_kind(methodHandle m)
                                 return java_lang_ref_reference_get;
   }
 
+  if (UseF2jBLASIntrinsics) {
+    switch (m->intrinsic_id()) {
+      case vmIntrinsics::_dgemm_dgemm: return org_netlib_blas_Dgemm_dgemm;
+      case vmIntrinsics::_dgemv_dgemv: return org_netlib_blas_Dgemv_dgemv;
+    }
+  }
+
   // Accessor method?
   if (m->is_accessor()) {
     assert(m->size_of_parameters() == 1, "fast code for accessors assumes parameter size = 1");
@@ -311,6 +330,8 @@ void AbstractInterpreter::print_method_kind(MethodKind kind) {
     case java_util_zip_CRC32_update           : tty->print("java_util_zip_CRC32_update"); break;
     case java_util_zip_CRC32_updateBytes      : tty->print("java_util_zip_CRC32_updateBytes"); break;
     case java_util_zip_CRC32_updateByteBuffer : tty->print("java_util_zip_CRC32_updateByteBuffer"); break;
+    case org_netlib_blas_Dgemm_dgemm : tty->print("org_netlib_blas_Dgemm_dgemm"); break;
+    case org_netlib_blas_Dgemv_dgemv : tty->print("org_netlib_blas_Dgemv_dgemv"); break;
     default:
       if (kind >= method_handle_invoke_FIRST &&
           kind <= method_handle_invoke_LAST) {
diff --git a/hotspot/src/share/vm/interpreter/templateInterpreter.cpp b/hotspot/src/share/vm/interpreter/templateInterpreter.cpp
index 1520c7b1c..f38f05117 100644
--- a/hotspot/src/share/vm/interpreter/templateInterpreter.cpp
+++ b/hotspot/src/share/vm/interpreter/templateInterpreter.cpp
@@ -32,12 +32,20 @@
 
 # define __ _masm->
 
-void TemplateInterpreter::initialize() {
+void TemplateInterpreter::initialize_stub() {
   if (_code != NULL) return;
   // assertions
   assert((int)Bytecodes::number_of_codes <= (int)DispatchTable::length,
          "dispatch table too small");
 
+  // allocate interpreter
+  int code_size = InterpreterCodeSize;
+  NOT_PRODUCT(code_size *= 4;)  // debug uses extra interpreter code space
+  _code = new StubQueue(new InterpreterCodeletInterface, code_size, NULL,
+                        "Interpreter");
+}
+
+void TemplateInterpreter::initialize_code() {
   AbstractInterpreter::initialize();
 
   TemplateTable::initialize();
@@ -45,10 +53,6 @@ void TemplateInterpreter::initialize() {
   // generate interpreter
   { ResourceMark rm;
     TraceTime timer("Interpreter generation", TraceStartupTime);
-    int code_size = InterpreterCodeSize;
-    NOT_PRODUCT(code_size *= 4;)  // debug uses extra interpreter code space
-    _code = new StubQueue(new InterpreterCodeletInterface, code_size, NULL,
-                          "Interpreter");
     InterpreterGenerator g(_code);
     if (PrintInterpreter) print();
   }
@@ -401,6 +405,11 @@ void TemplateInterpreterGenerator::generate_all() {
     method_entry(java_util_zip_CRC32_updateByteBuffer)
   }
 
+  if (UseF2jBLASIntrinsics) {
+    method_entry(org_netlib_blas_Dgemm_dgemm)
+    method_entry(org_netlib_blas_Dgemv_dgemv)
+  }
+
   initialize_method_handle_entries();
 
   // all native method kinds (must be one contiguous block)
diff --git a/hotspot/src/share/vm/interpreter/templateInterpreter.hpp b/hotspot/src/share/vm/interpreter/templateInterpreter.hpp
index 5f76dca8a..96da6353c 100644
--- a/hotspot/src/share/vm/interpreter/templateInterpreter.hpp
+++ b/hotspot/src/share/vm/interpreter/templateInterpreter.hpp
@@ -132,7 +132,8 @@ class TemplateInterpreter: public AbstractInterpreter {
 
  public:
   // Initialization/debugging
-  static void       initialize();
+  static void       initialize_stub();
+  static void       initialize_code();
   // this only returns whether a pc is within generated code for the interpreter.
   static bool       contains(address pc)                        { return _code != NULL && _code->contains(pc); }
 
diff --git a/hotspot/src/share/vm/opto/escape.cpp b/hotspot/src/share/vm/opto/escape.cpp
index 68631dbf2..0e0cc1028 100644
--- a/hotspot/src/share/vm/opto/escape.cpp
+++ b/hotspot/src/share/vm/opto/escape.cpp
@@ -979,7 +979,9 @@ void ConnectionGraph::process_call_arguments(CallNode *call) {
                   strcmp(call->as_CallLeaf()->_name, "mulAdd") == 0 ||
                   strcmp(call->as_CallLeaf()->_name, "montgomery_multiply") == 0 ||
                   strcmp(call->as_CallLeaf()->_name, "montgomery_square") == 0 ||
-                  strcmp(call->as_CallLeaf()->_name, "f2jblas_ddot") == 0)
+                  strcmp(call->as_CallLeaf()->_name, "f2jblas_ddot") == 0 ||
+                  strcmp(call->as_CallLeaf()->_name, "dgemm_dgemm") == 0) ||
+                  strcmp(call->as_CallLeaf()->_name, "dgemv_dgemv") == 0
                  ))) {
             call->dump();
             fatal(err_msg_res("EA unexpected CallLeaf %s", call->as_CallLeaf()->_name));
diff --git a/hotspot/src/share/vm/opto/graphKit.cpp b/hotspot/src/share/vm/opto/graphKit.cpp
index 41a067ce2..1c3bc2e8c 100644
--- a/hotspot/src/share/vm/opto/graphKit.cpp
+++ b/hotspot/src/share/vm/opto/graphKit.cpp
@@ -2372,7 +2372,11 @@ Node* GraphKit::make_runtime_call(int flags,
                                   Node* parm0, Node* parm1,
                                   Node* parm2, Node* parm3,
                                   Node* parm4, Node* parm5,
-                                  Node* parm6, Node* parm7) {
+                                  Node* parm6, Node* parm7,
+                                  Node* parm8, Node* parm9,
+                                  Node* parm10, Node* parm11,
+                                  Node* parm12, Node* parm13,
+                                  Node* parm14, Node* parm15) {
   // Slow-path call
   bool is_leaf = !(flags & RC_NO_LEAF);
   bool has_io  = (!is_leaf && !(flags & RC_NO_IO));
@@ -2415,7 +2419,15 @@ Node* GraphKit::make_runtime_call(int flags,
   if (parm5 != NULL) { call->init_req(TypeFunc::Parms+5, parm5);
   if (parm6 != NULL) { call->init_req(TypeFunc::Parms+6, parm6);
   if (parm7 != NULL) { call->init_req(TypeFunc::Parms+7, parm7);
-    /* close each nested if ===> */  } } } } } } } }
+  if (parm8 != NULL) { call->init_req(TypeFunc::Parms+8, parm8);
+  if (parm9 != NULL) { call->init_req(TypeFunc::Parms+9, parm9);
+  if (parm10 != NULL) { call->init_req(TypeFunc::Parms+10, parm10);
+  if (parm11 != NULL) { call->init_req(TypeFunc::Parms+11, parm11);
+  if (parm12 != NULL) { call->init_req(TypeFunc::Parms+12, parm12);
+  if (parm13 != NULL) { call->init_req(TypeFunc::Parms+13, parm13);
+  if (parm14 != NULL) { call->init_req(TypeFunc::Parms+14, parm14);
+  if (parm15 != NULL) { call->init_req(TypeFunc::Parms+15, parm15);
+    /* close each nested if ===> */  } } } } } } } } } } } } } } } }
   assert(call->in(call->req()-1) != NULL, "must initialize all parms");
 
   if (!is_leaf) {
diff --git a/hotspot/src/share/vm/opto/graphKit.hpp b/hotspot/src/share/vm/opto/graphKit.hpp
index 7a363fd33..e9a061acf 100644
--- a/hotspot/src/share/vm/opto/graphKit.hpp
+++ b/hotspot/src/share/vm/opto/graphKit.hpp
@@ -818,7 +818,11 @@ class GraphKit : public Phase {
                           Node* parm0 = NULL, Node* parm1 = NULL,
                           Node* parm2 = NULL, Node* parm3 = NULL,
                           Node* parm4 = NULL, Node* parm5 = NULL,
-                          Node* parm6 = NULL, Node* parm7 = NULL);
+                          Node* parm6 = NULL, Node* parm7 = NULL,
+                          Node* parm8 = NULL, Node* parm9 = NULL,
+                          Node* parm10 = NULL, Node* parm11 = NULL,
+                          Node* parm12 = NULL, Node* parm13 = NULL,
+                          Node* parm14 = NULL, Node* parm15 = NULL);
   enum {  // flag values for make_runtime_call
     RC_NO_FP = 1,               // CallLeafNoFPNode
     RC_NO_IO = 2,               // do not hook IO edges
diff --git a/hotspot/src/share/vm/opto/library_call.cpp b/hotspot/src/share/vm/opto/library_call.cpp
index 5cbc0f012..10eeea217 100644
--- a/hotspot/src/share/vm/opto/library_call.cpp
+++ b/hotspot/src/share/vm/opto/library_call.cpp
@@ -336,6 +336,8 @@ class LibraryCallKit : public GraphKit {
   bool inline_montgomeryMultiply();
   bool inline_montgomerySquare();
   bool inline_ddotF2jBLAS();
+  bool inline_dgemmDgemm();
+  bool inline_dgemvDgemv();
 
   bool inline_profileBoolean();
 };
@@ -589,6 +591,8 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
     break;
 
   case vmIntrinsics::_f2jblas_ddot:
+  case vmIntrinsics::_dgemm_dgemm:
+  case vmIntrinsics::_dgemv_dgemv:
     if (!UseF2jBLASIntrinsics) return NULL;
     break;
 
@@ -988,9 +992,13 @@ bool LibraryCallKit::try_to_inline(int predicate) {
 
   case vmIntrinsics::_profileBoolean:
     return inline_profileBoolean();
+
   case vmIntrinsics::_f2jblas_ddot:
     return inline_ddotF2jBLAS();
-
+  case vmIntrinsics::_dgemm_dgemm:
+    return inline_dgemmDgemm();
+  case vmIntrinsics::_dgemv_dgemv:
+    return inline_dgemvDgemv();
   default:
     // If you get here, it may be that someone has added a new intrinsic
     // to the list in vmSymbols.hpp without implementing it here.
@@ -6354,6 +6362,144 @@ bool LibraryCallKit::inline_ddotF2jBLAS() {
 }
 
 /**
+ * double org.netlib.blas.Dgemm.dgemm(java.lang.String transa,
+ *                                    java.lang.String transb, int m, int n, int k,
+ *                                    double alpha, double[] a, int offset_a, int lda,
+ *                                    double[] b, int offset_b, int ldb, double beta,
+ *                                    double[] c, int offset_c, int Ldc)
+ */
+bool LibraryCallKit::inline_dgemmDgemm() {
+  assert(callee()->signature()->count() == 16, "Dgemm.dgemm has 16 parameters");
+
+  address stubAddr = StubRoutines::dgemmDgemm();
+  if (stubAddr == NULL) return false;
+
+  Node* transa = argument(0);
+  Node* transb = argument(1);
+  Node* m = argument(2);
+  Node* n = argument(3);
+  Node* k = argument(4);
+  Node* alpha = round_double_node(argument(5));
+  Node* a = argument(7);
+  Node* a_offset = argument(8);
+  Node* lda = argument(9);
+  Node* b = argument(10);
+  Node* b_offset = argument(11);
+  Node* ldb = argument(12);
+  Node* beta = round_double_node(argument(13));
+  Node* c = argument(15);
+  Node* c_offset = argument(16);
+  Node* ldc = argument(17);
+
+  const Type* a_type = a->Value(&_gvn);
+  const Type* b_type = b->Value(&_gvn);
+  const Type* c_type = c->Value(&_gvn);
+  const TypeAryPtr* a_base_type = a_type->isa_aryptr();
+  const TypeAryPtr* b_base_type = b_type->isa_aryptr();
+  const TypeAryPtr* c_base_type = c_type->isa_aryptr();
+  if (a_base_type == NULL || b_base_type == NULL || c_base_type == NULL) return false;
+
+  ciKlass* a_klass = a_base_type->klass();
+  ciKlass* b_klass = b_base_type->klass();
+  ciKlass* c_klass = c_base_type->klass();
+  if (a_klass == NULL || b_klass == NULL || c_klass == NULL) return false;
+
+  BasicType a_elem_type = a_klass->as_array_klass()->element_type()->basic_type();
+  BasicType b_elem_type = b_klass->as_array_klass()->element_type()->basic_type();
+  BasicType c_elem_type = a_klass->as_array_klass()->element_type()->basic_type();
+  if (a_elem_type != T_DOUBLE || b_elem_type != T_DOUBLE || c_elem_type != T_DOUBLE) return false;
+
+  // get array a/b/c's addr
+  Node* a_start = array_element_address(a, a_offset, a_elem_type);
+  Node* b_start = array_element_address(b, b_offset, b_elem_type);
+  Node* c_start = array_element_address(c, c_offset, c_elem_type);
+
+  // Get start addr of string
+  Node* transa_value   = load_String_value(NULL, transa);
+  Node* transa_offset  = load_String_offset(NULL, transa);
+  Node* transa_start   = array_element_address(transa_value, transa_offset, T_CHAR);
+  Node* transb_value   = load_String_value(NULL, transb);
+  Node* transb_offset  = load_String_offset(NULL, transb);
+  Node* transb_start   = array_element_address(transb_value, transb_offset, T_CHAR);
+
+  const char *stubName = "dgemm_dgemm";
+  make_runtime_call(RC_LEAF, OptoRuntime::dgemmDgemm_Type(),
+                            stubAddr, stubName, TypePtr::BOTTOM,
+                            transa_start, transb_start, m, n, k, alpha, top(),
+                            a_start, lda, b_start, ldb, beta, top(), c_start, ldc);
+
+  return true;
+}
+
+/**
+ * void org.netlib.blas.Dgemv.dgemv(string trans, int m, int n, double alpha,
+ *                                  double[] a, int _a_offset, int lda,
+ *                                  double[] x, int _x_offset, int incx, double beta,
+ *                                  double[] y, int _y_offset, int incy)
+ */
+bool LibraryCallKit::inline_dgemvDgemv() {
+  assert(callee()->signature()->count() == 14, "F2jBLAS.dgemv has 14 parameters");
+  Node* trans = argument(0);
+  Node* m = argument(1);
+  Node* n = argument(2);
+  Node* alpha = round_double_node(argument(3));
+  Node* a = argument(5);
+  Node* a_offset = argument(6);
+  Node* lda = argument(7);
+  Node* x = argument(8);
+  Node* x_offset = argument(9);
+  Node* incx = argument(10);
+  Node* beta = round_double_node(argument(11));
+  Node* y = argument(13);
+  Node* y_offset = argument(14);
+  Node* incy = argument(15);
+
+  const Type* a_type = a->Value(&_gvn);
+  const Type* x_type = x->Value(&_gvn);
+  const Type* y_type = y->Value(&_gvn);
+  const TypeAryPtr* a_base_type = a_type->isa_aryptr();
+  const TypeAryPtr* x_base_type = x_type->isa_aryptr();
+  const TypeAryPtr* y_base_type = y_type->isa_aryptr();
+  if (a_base_type == NULL || x_base_type == NULL || y_base_type == NULL) return false;
+
+  ciKlass* a_klass = a_base_type->klass();
+  ciKlass* x_klass = x_base_type->klass();
+  ciKlass* y_klass = y_base_type->klass();
+
+  if (a_klass == NULL || x_klass == NULL || y_klass == NULL) return false;
+
+  BasicType a_elem_type = a_klass->as_array_klass()->element_type()->basic_type();
+  BasicType x_elem_type = x_klass->as_array_klass()->element_type()->basic_type();
+  BasicType y_elem_type = y_klass->as_array_klass()->element_type()->basic_type();
+
+  if (a_elem_type != T_DOUBLE || x_elem_type != T_DOUBLE || y_elem_type != T_DOUBLE) return false;
+
+
+  address stubAddr = StubRoutines::dgemvDgemv();
+  if (stubAddr == NULL) return false;
+
+  // 'a_start' points to array a + scaled offset
+  Node* a_start = array_element_address(a, a_offset, a_elem_type);
+  // 'x_start' points to array x + scaled offset
+  Node* x_start = array_element_address(x, x_offset, x_elem_type);
+  // 'y_start' points to array y + scaled offset
+  Node* y_start = array_element_address(y, y_offset, y_elem_type);
+
+  Node* no_ctrl = NULL;
+
+  // get start addr of string
+  Node* trans_value   = load_String_value(no_ctrl, trans);
+  Node* trans_offset  = load_String_offset(no_ctrl, trans);
+  Node* trans_start   = array_element_address(trans_value, trans_offset, T_CHAR);
+
+  const char *stubName = "dgemv_dgemv";
+  Node* call = make_runtime_call(RC_LEAF, OptoRuntime::dgemvDgemv_Type(), stubAddr, stubName,
+                                 TypePtr::BOTTOM, trans_start, m, n, alpha, top(), a_start,
+                                 lda, x_start, incx, beta, top(), y_start, incy);
+  return true;
+}
+
+/**
  * Calculate CRC32 for ByteBuffer.
  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
  */
diff --git a/hotspot/src/share/vm/opto/runtime.cpp b/hotspot/src/share/vm/opto/runtime.cpp
index f1fe4d666..dc8f0c774 100644
--- a/hotspot/src/share/vm/opto/runtime.cpp
+++ b/hotspot/src/share/vm/opto/runtime.cpp
@@ -944,6 +944,81 @@ const TypeFunc* OptoRuntime::ddotF2jBLAS_Type() {
   return TypeFunc::make(domain, range);
 }
 
+/**
+ * double org.netlib.blas.Dgemm.dgemm(java.lang.String transa,
+ *                                    java.lang.String transb, int m, int n, int k,
+ *                                    double alpha, double[] a, int offset_a, int lda,
+ *                                    double[] b, int offset_b, int ldb, double beta,
+ *                                    double[] c, int offset_c, int Ldc)
+ */
+const TypeFunc* OptoRuntime::dgemmDgemm_Type() {
+    // create input type (domain)
+    int num_args = 15;
+    int argcnt = num_args;
+    const Type** fields = TypeTuple::fields(argcnt);
+    int argp = TypeFunc::Parms;
+
+    fields[argp++] = TypeAryPtr::CHARS;       // char[]
+    fields[argp++] = TypeAryPtr::CHARS;       // char[]
+    fields[argp++] = TypeInt::INT;            // int m
+    fields[argp++] = TypeInt::INT;            // int n
+    fields[argp++] = TypeInt::INT;            // int k
+    fields[argp++] = Type::DOUBLE;            // double alpha
+    fields[argp++] = Type::HALF;
+    fields[argp++] = TypeAryPtr::DOUBLES;     // double[] a
+    fields[argp++] = TypeInt::INT;            // int lda
+    fields[argp++] = TypeAryPtr::DOUBLES;     // double[] b
+    fields[argp++] = TypeInt::INT;            // int ldb
+    fields[argp++] = Type::DOUBLE;            // double beta
+    fields[argp++] = Type::HALF;
+    fields[argp++] = TypeAryPtr::DOUBLES;     // double[] c
+    fields[argp++] = TypeInt::INT;            // int ldc
+    assert(argp == TypeFunc::Parms + argcnt, "correct decoding");
+    const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms + argcnt, fields);
+
+    // no result type needed
+    fields = TypeTuple::fields(1);
+    fields[TypeFunc::Parms + 0] = NULL;       // void
+    const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
+    return TypeFunc::make(domain, range);
+}
+
+/**
+ * void dgemv(String trans, int m, int n, double alpha,
+ *            double[] a, int _a_offset, int lda,
+ *            double[] x, int _x_offset, int incx, double beta,
+ *            double[] y, int _y_offset, int incy)
+ */
+const TypeFunc* OptoRuntime::dgemvDgemv_Type() {
+  // create input type (domain)
+  int num_args = 13;
+  int argcnt = num_args;
+  const Type** fields = TypeTuple::fields(argcnt);
+  int argp = TypeFunc::Parms;
+
+  fields[argp++] = TypeAryPtr::CHARS;       // char[]
+  fields[argp++] = TypeInt::INT;            // int m
+  fields[argp++] = TypeInt::INT;            // int n
+  fields[argp++] = Type::DOUBLE;            // double alpha
+  fields[argp++] = Type::HALF;
+  fields[argp++] = TypeAryPtr::DOUBLES;     // double[] a
+  fields[argp++] = TypeInt::INT;            // int lda
+  fields[argp++] = TypeAryPtr::DOUBLES;     // double[] x
+  fields[argp++] = TypeInt::INT;            // int incx
+  fields[argp++] = Type::DOUBLE;            // double beta
+  fields[argp++] = Type::HALF;
+  fields[argp++] = TypeAryPtr::DOUBLES;     // double[] y
+  fields[argp++] = TypeInt::INT;            // int incy
+  assert(argp == TypeFunc::Parms + argcnt, "correct decoding");
+  const TypeTuple* domain = TypeTuple::make(TypeFunc::Parms + argcnt, fields);
+
+  // no result type needed
+  fields = TypeTuple::fields(1);
+  fields[TypeFunc::Parms + 0] = NULL;       // void
+  const TypeTuple* range = TypeTuple::make(TypeFunc::Parms, fields);
+  return TypeFunc::make(domain, range);
+}
+
 // for cipherBlockChaining calls of aescrypt encrypt/decrypt, four pointers and a length, returning int
 const TypeFunc* OptoRuntime::cipherBlockChaining_aescrypt_Type() {
   // create input type (domain)
diff --git a/hotspot/src/share/vm/opto/runtime.hpp b/hotspot/src/share/vm/opto/runtime.hpp
index 66d393c5c..e07c34c15 100644
--- a/hotspot/src/share/vm/opto/runtime.hpp
+++ b/hotspot/src/share/vm/opto/runtime.hpp
@@ -318,6 +318,8 @@ private:
   static const TypeFunc* updateBytesCRC32_Type();
 
   static const TypeFunc* ddotF2jBLAS_Type();
+  static const TypeFunc* dgemmDgemm_Type();
+  static const TypeFunc* dgemvDgemv_Type();
 
   // leaf on stack replacement interpreter accessor types
   static const TypeFunc* osr_end_Type();
diff --git a/hotspot/src/share/vm/runtime/init.cpp b/hotspot/src/share/vm/runtime/init.cpp
index 1512ccc96..4c133bd4e 100644
--- a/hotspot/src/share/vm/runtime/init.cpp
+++ b/hotspot/src/share/vm/runtime/init.cpp
@@ -54,7 +54,8 @@ void VM_Version_init();
 void os_init_globals();        // depends on VM_Version_init, before universe_init
 void stubRoutines_init1();
 jint universe_init();          // depends on codeCache_init and stubRoutines_init
-void interpreter_init();       // before any methods loaded
+void interpreter_init_stub();  // before any methods loaded
+void interpreter_init_code();  // after methods loaded, but before they are linked
 void invocationCounter_init(); // before any methods loaded
 void marksweep_init();
 void accessFlags_init();
@@ -106,7 +107,7 @@ jint init_globals() {
   if (status != JNI_OK)
     return status;
 
-  interpreter_init();  // before any methods loaded
+  interpreter_init_stub(); // before methods get loaded
   invocationCounter_init();  // before any methods loaded
   marksweep_init();
   accessFlags_init();
@@ -114,6 +115,7 @@ jint init_globals() {
   InterfaceSupport_init();
   SharedRuntime::generate_stubs();
   universe2_init();  // dependent on codeCache_init and stubRoutines_init1
+  interpreter_init_code(); // after universe2_init and before any method gets linked
   referenceProcessor_init();
   jni_handles_init();
 #if INCLUDE_VM_STRUCTS
diff --git a/hotspot/src/share/vm/runtime/stubRoutines.cpp b/hotspot/src/share/vm/runtime/stubRoutines.cpp
index 10f438bc5..f2106d13a 100644
--- a/hotspot/src/share/vm/runtime/stubRoutines.cpp
+++ b/hotspot/src/share/vm/runtime/stubRoutines.cpp
@@ -136,7 +136,10 @@ address StubRoutines::_sha512_implCompressMB = NULL;
 address StubRoutines::_updateBytesCRC32 = NULL;
 address StubRoutines::_crc_table_adr = NULL;
 
+address StubRoutines::_BLAS_library = NULL;
 address StubRoutines::_ddotF2jBLAS = NULL;
+address StubRoutines::_dgemmDgemm = NULL;
+address StubRoutines::_dgemvDgemv = NULL;
 
 address StubRoutines::_multiplyToLen = NULL;
 address StubRoutines::_squareToLen = NULL;
diff --git a/hotspot/src/share/vm/runtime/stubRoutines.hpp b/hotspot/src/share/vm/runtime/stubRoutines.hpp
index a4eeb910d..16075d9f4 100644
--- a/hotspot/src/share/vm/runtime/stubRoutines.hpp
+++ b/hotspot/src/share/vm/runtime/stubRoutines.hpp
@@ -214,7 +214,10 @@ class StubRoutines: AllStatic {
   static address _updateBytesCRC32;
   static address _crc_table_adr;
 
+  static address _BLAS_library;
   static address _ddotF2jBLAS;
+  static address _dgemmDgemm;
+  static address _dgemvDgemv;
 
   static address _multiplyToLen;
   static address _squareToLen;
@@ -380,6 +383,8 @@ class StubRoutines: AllStatic {
   static address crc_table_addr()      { return _crc_table_adr; }
 
   static address ddotF2jBLAS()         { return _ddotF2jBLAS; }
+  static address dgemmDgemm()          { return _dgemmDgemm; }
+  static address dgemvDgemv()          { return _dgemvDgemv; }
 
   static address multiplyToLen()       {return _multiplyToLen; }
   static address squareToLen()         {return _squareToLen; }
-- 
2.12.3