8269404: Base64 Encoding optimization enhancements for x86 using AVX-512

Reviewed-by: kvn, sviswanathan

src/hotspot/cpu/x86/assembler_x86.cpp

@@ -3757,6 +3757,15 @@ void Assembler::vpermb(XMMRegister dst, XMMRegister nds, XMMRegister src, int ve
   emit_int16((unsigned char)0x8D, (0xC0 | encode));
 }
 
+void Assembler::vpermb(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
+  assert(VM_Version::supports_avx512_vbmi(), "");
+  InstructionAttr attributes(vector_len, /* rex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ true);
+  attributes.set_is_evex_instruction();
+  vex_prefix(src, nds->encoding(), dst->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int8((unsigned char)0x8D);
+  emit_operand(dst, src);
+}
+
 void Assembler::vpermw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
   assert(vector_len == AVX_128bit ? VM_Version::supports_avx512vlbw() :
          vector_len == AVX_256bit ? VM_Version::supports_avx512vlbw() :
@@ -3837,6 +3846,14 @@ void Assembler::evpermt2b(XMMRegister dst, XMMRegister nds, XMMRegister src, int
   emit_int16(0x7D, (0xC0 | encode));
 }
 
+void Assembler::evpmultishiftqb(XMMRegister dst, XMMRegister ctl, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_avx512_vbmi(), "");
+  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ true);
+  attributes.set_is_evex_instruction();
+  int encode = vex_prefix_and_encode(dst->encoding(), ctl->encoding(), src->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int16((unsigned char)0x83, (unsigned char)(0xC0 | encode));
+}
+
 void Assembler::pause() {
   emit_int16((unsigned char)0xF3, (unsigned char)0x90);
 }
@@ -4135,6 +4152,15 @@ void Assembler::vpmovmskb(Register dst, XMMRegister src, int vec_enc) {
   emit_int16((unsigned char)0xD7, (0xC0 | encode));
 }
 
+void Assembler::vpmaskmovd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
+  assert((VM_Version::supports_avx2() && vector_len == AVX_256bit), "");
+  InstructionMark im(this);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ false, /* uses_vl */ true);
+  vex_prefix(src, nds->encoding(), dst->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int8((unsigned char)0x8C);
+  emit_operand(dst, src);
+}
+
 void Assembler::pextrd(Register dst, XMMRegister src, int imm8) {
   assert(VM_Version::supports_sse4_1(), "");
   InstructionAttr attributes(AVX_128bit, /* rex_w */ false, /* legacy_mode */ _legacy_mode_dq, /* no_mask_reg */ true, /* uses_vl */ false);
@@ -6564,6 +6590,13 @@ void Assembler::psubq(XMMRegister dst, XMMRegister src) {
   emit_int8((0xC0 | encode));
 }
 
+void Assembler::vpsubusb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
+  assert(UseAVX > 0, "requires some form of AVX");
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
+  int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_66, VEX_OPCODE_0F, &attributes);
+  emit_int16((unsigned char)0xD8, (0xC0 | encode));
+}
+
 void Assembler::vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
   assert(UseAVX > 0, "requires some form of AVX");
   InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
@@ -6655,6 +6688,15 @@ void Assembler::pmuludq(XMMRegister dst, XMMRegister src) {
   emit_int16((unsigned char)0xF4, (0xC0 | encode));
 }
 
+void Assembler::vpmulhuw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
+  assert((vector_len == AVX_128bit && VM_Version::supports_avx()) ||
+         (vector_len == AVX_256bit && VM_Version::supports_avx2()) ||
+         (vector_len == AVX_512bit && VM_Version::supports_avx512bw()), "");
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
+  int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_66, VEX_OPCODE_0F, &attributes);
+  emit_int16((unsigned char)0xE4, (0xC0 | encode));
+}
+
 void Assembler::vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
   assert(UseAVX > 0, "requires some form of AVX");
   InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);

src/hotspot/cpu/x86/assembler_x86.hpp

@@ -1688,6 +1688,7 @@ class Assembler : public AbstractAssembler  {
   void vpermq(XMMRegister dst, XMMRegister src, int imm8);
   void vpermq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
   void vpermb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
+  void vpermb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
   void vpermw(XMMRegister dst,  XMMRegister nds, XMMRegister src, int vector_len);
   void vpermd(XMMRegister dst,  XMMRegister nds, Address src, int vector_len);
   void vpermd(XMMRegister dst,  XMMRegister nds, XMMRegister src, int vector_len);
@@ -1698,6 +1699,7 @@ class Assembler : public AbstractAssembler  {
   void vpermpd(XMMRegister dst, XMMRegister src, int imm8, int vector_len);
   void evpermi2q(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
   void evpermt2b(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
+  void evpmultishiftqb(XMMRegister dst, XMMRegister ctl, XMMRegister src, int vector_len);
 
   void pause();
 
@@ -1746,6 +1748,7 @@ class Assembler : public AbstractAssembler  {
 
   void pmovmskb(Register dst, XMMRegister src);
   void vpmovmskb(Register dst, XMMRegister src, int vec_enc);
+  void vpmaskmovd(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
 
   // SSE 4.1 extract
   void pextrd(Register dst, XMMRegister src, int imm8);
@@ -2248,6 +2251,7 @@ class Assembler : public AbstractAssembler  {
   void psubw(XMMRegister dst, XMMRegister src);
   void psubd(XMMRegister dst, XMMRegister src);
   void psubq(XMMRegister dst, XMMRegister src);
+  void vpsubusb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
   void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
   void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
   void vpsubd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
@@ -2268,6 +2272,7 @@ class Assembler : public AbstractAssembler  {
   void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
   void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
   void vpmullq(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
+  void vpmulhuw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
 
   // Minimum of packed integers
   void pminsb(XMMRegister dst, XMMRegister src);