src/cpu/x86/vm/assembler_x86.cpp

@@ -2321,10 +2321,19 @@
   int encode = vex_prefix_and_encode(src1->encoding(), 0, src2->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F, &attributes);
   emit_int8((unsigned char)0x98);
   emit_int8((unsigned char)(0xC0 | encode));
 }
 
+// This instruction produces ZF or CF flags
+void Assembler::ktestql(KRegister src1, KRegister src2) {
+  assert(VM_Version::supports_avx512bw(), "");
+  InstructionAttr attributes(AVX_128bit, /* rex_w */ true, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ false);
+  int encode = vex_prefix_and_encode(src1->encoding(), 0, src2->encoding(), VEX_SIMD_NONE, VEX_OPCODE_0F, &attributes);
+  emit_int8((unsigned char)0x99);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
 void Assembler::movb(Address dst, int imm8) {
   InstructionMark im(this);
    prefix(dst);
   emit_int8((unsigned char)0xC6);
   emit_operand(rax, dst, 1);

@@ -2489,10 +2498,23 @@
   vex_prefix(dst, 0, src->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
   emit_int8(0x7F);
   emit_operand(src, dst);
 }
 
+void Assembler::evmovdqub(KRegister mask, XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_avx512vlbw(), "");
+  assert(is_vector_masking(), "");    // For stub code use only
+  InstructionMark im(this);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
+  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
+  attributes.set_embedded_opmask_register_specifier(mask);
+  attributes.set_is_evex_instruction();
+  vex_prefix(src, 0, dst->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
+  emit_int8(0x6F);
+  emit_operand(dst, src);
+}
+
 void Assembler::evmovdquw(XMMRegister dst, XMMRegister src, int vector_len) {
   assert(VM_Version::supports_evex(), "");
   InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
   attributes.set_is_evex_instruction();
   int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;

@@ -3273,10 +3295,23 @@
   vex_prefix(src, nds->encoding(), dst_enc, VEX_SIMD_66, VEX_OPCODE_0F, &attributes);
   emit_int8(0x74);
   emit_operand(as_Register(dst_enc), src);
 }
 
+void Assembler::evpcmpeqb(KRegister mask, KRegister kdst, XMMRegister nds, Address src, int vector_len) {
+  assert(VM_Version::supports_avx512vlbw(), "");
+  assert(is_vector_masking(), "");    // For stub code use only
+  InstructionMark im(this);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ false, /* no_reg_mask */ false, /* uses_vl */ false);
+  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
+  attributes.set_embedded_opmask_register_specifier(mask);
+  attributes.set_is_evex_instruction();
+  vex_prefix(src, nds->encoding(), kdst->encoding(), VEX_SIMD_66, VEX_OPCODE_0F, &attributes);
+  emit_int8(0x74);
+  emit_operand(as_Register(kdst->encoding()), src);
+}
+
 // In this context, the dst vector contains the components that are equal, non equal components are zeroed in dst
 void Assembler::pcmpeqw(XMMRegister dst, XMMRegister src) {
   assert(VM_Version::supports_sse2(), "");
   InstructionAttr attributes(AVX_128bit, /* rex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ false, /* uses_vl */ false);
   int encode = simd_prefix_and_encode(dst, dst, src, VEX_SIMD_66, VEX_OPCODE_0F, &attributes);

@@ -6894,11 +6929,11 @@
   // of form {66, F3, F2}
   byte3 |= pre;
   emit_int8(byte3);
 
   // P2: byte 4 as zL'Lbv'aaa
-  int byte4 = (_attributes->is_no_reg_mask()) ? 0 : 1; // kregs are implemented in the low 3 bits as aaa (hard code k1, it will be initialized for now)
+  int byte4 = (_attributes->is_no_reg_mask()) ? 0 : _attributes->get_embedded_opmask_register_specifier(); // kregs are implemented in the low 3 bits as aaa (hard code k1, it will be initialized for now)
   // EVEX.v` for extending EVEX.vvvv or VIDX
   byte4 |= (evex_v ? 0: EVEX_V);
   // third EXEC.b for broadcast actions
   byte4 |= (_attributes->is_extended_context() ? EVEX_Rb : 0);
   // fourth EVEX.L'L for vector length : 0 is 128, 1 is 256, 2 is 512, currently we do not support 1024