8268276: Base64 Decoding optimization for x86 using AVX-512

Reviewed-by: erikj, sviswanathan, kvn

Author: asgibbons

.gitignore

@@ -16,3 +16,5 @@ NashornProfile.txt
 **/JTreport/**
 **/JTwork/**
 /src/utils/LogCompilation/target/
+/.project/
+/.settings/

src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp

@@ -5683,6 +5683,7 @@ class StubGenerator: public StubCodeGenerator {
    *  c_rarg3   - dest_start
    *  c_rarg4   - dest_offset
    *  c_rarg5   - isURL
+   *  c_rarg6   - isMIME
    *
    */
   address generate_base64_decodeBlock() {
@@ -5765,12 +5766,13 @@ class StubGenerator: public StubCodeGenerator {
     StubCodeMark mark(this, "StubRoutines", "decodeBlock");
     address start = __ pc();
 
-    Register src   = c_rarg0;  // source array
-    Register soff  = c_rarg1;  // source start offset
-    Register send  = c_rarg2;  // source end offset
-    Register dst   = c_rarg3;  // dest array
-    Register doff  = c_rarg4;  // position for writing to dest array
-    Register isURL = c_rarg5;  // Base64 or URL character set
+    Register src    = c_rarg0;  // source array
+    Register soff   = c_rarg1;  // source start offset
+    Register send   = c_rarg2;  // source end offset
+    Register dst    = c_rarg3;  // dest array
+    Register doff   = c_rarg4;  // position for writing to dest array
+    Register isURL  = c_rarg5;  // Base64 or URL character set
+    Register isMIME = c_rarg6;  // Decoding MIME block - unused in this implementation
 
     Register length = send;    // reuse send as length of source data to process
 

src/hotspot/cpu/ppc/stubGenerator_ppc.cpp

@@ -3757,6 +3757,7 @@ class StubGenerator: public StubCodeGenerator {
     Register d      = R6_ARG4; // destination address
     Register dp     = R7_ARG5; // destination offset
     Register isURL  = R8_ARG6; // boolean, if non-zero indicates use of RFC 4648 base64url encoding
+    Register isMIME = R9_ARG7; // boolean, if non-zero indicates use of RFC 2045 MIME encoding - not used
 
     // Local variables
     Register const_ptr     = R9;  // used for loading constants

src/hotspot/cpu/x86/assembler_x86.cpp

@@ -3829,6 +3829,14 @@ void Assembler::evpermi2q(XMMRegister dst, XMMRegister nds, XMMRegister src, int
   emit_int16(0x76, (0xC0 | encode));
 }
 
+void Assembler::evpermt2b(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_avx512_vbmi(), "");
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ true);
+  attributes.set_is_evex_instruction();
+  int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int16(0x7D, (0xC0 | encode));
+}
+
 void Assembler::pause() {
   emit_int16((unsigned char)0xF3, (unsigned char)0x90);
 }
@@ -4548,6 +4556,15 @@ void Assembler::vpmaddwd(XMMRegister dst, XMMRegister nds, XMMRegister src, int
   emit_int16((unsigned char)0xF5, (0xC0 | encode));
 }
 
+void Assembler::vpmaddubsw(XMMRegister dst, XMMRegister src1, XMMRegister src2, 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() : 0, "");
+  InstructionAttr attributes(vector_len, /* rex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
+  int encode = simd_prefix_and_encode(dst, src1, src2, VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int16(0x04, (0xC0 | encode));
+}
+
 void Assembler::evpdpwssd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
   assert(VM_Version::supports_evex(), "");
   assert(VM_Version::supports_avx512_vnni(), "must support vnni");
@@ -4856,6 +4873,15 @@ void Assembler::vptest(XMMRegister dst, XMMRegister src, int vector_len) {
   emit_int16(0x17, (0xC0 | encode));
 }
 
+void Assembler::evptestmb(KRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_avx512vlbw(), "");
+  // Encoding: EVEX.NDS.XXX.66.0F.W0 DB /r
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ true);
+  attributes.set_is_evex_instruction();
+  int encode = vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(), VEX_SIMD_66, VEX_OPCODE_0F_38, &attributes);
+  emit_int16((unsigned char)0x26, (0xC0 | encode));
+}
+
 void Assembler::punpcklbw(XMMRegister dst, Address src) {
   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
   assert((UseAVX > 0), "SSE mode requires address alignment 16 bytes");
@@ -9403,6 +9429,13 @@ void Assembler::shlxq(Register dst, Register src1, Register src2) {
   emit_int16((unsigned char)0xF7, (0xC0 | encode));
 }
 
+void Assembler::shrxl(Register dst, Register src1, Register src2) {
+  assert(VM_Version::supports_bmi2(), "");
+  InstructionAttr attributes(AVX_128bit, /* vex_w */ false, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);
+  int encode = vex_prefix_and_encode(dst->encoding(), src2->encoding(), src1->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F_38, &attributes);
+  emit_int16((unsigned char)0xF7, (0xC0 | encode));
+}
+
 void Assembler::shrxq(Register dst, Register src1, Register src2) {
   assert(VM_Version::supports_bmi2(), "");
   InstructionAttr attributes(AVX_128bit, /* vex_w */ true, /* legacy_mode */ true, /* no_mask_reg */ true, /* uses_vl */ true);

src/hotspot/cpu/x86/assembler_x86.hpp

@@ -1697,6 +1697,7 @@ class Assembler : public AbstractAssembler  {
   void vpermilpd(XMMRegister dst, XMMRegister src, int imm8, int vector_len);
   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 pause();
 
@@ -1810,6 +1811,8 @@ class Assembler : public AbstractAssembler  {
   // Multiply add
   void pmaddwd(XMMRegister dst, XMMRegister src);
   void vpmaddwd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
+  void vpmaddubsw(XMMRegister dst, XMMRegister src1, XMMRegister src2, int vector_len);
+
   // Multiply add accumulate
   void evpdpwssd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
 
@@ -1877,6 +1880,8 @@ class Assembler : public AbstractAssembler  {
   void vptest(XMMRegister dst, XMMRegister src);
   void vptest(XMMRegister dst, Address src);
 
+  void evptestmb(KRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
+
   // Vector compare
   void vptest(XMMRegister dst, XMMRegister src, int vector_len);
 
@@ -2138,6 +2143,7 @@ class Assembler : public AbstractAssembler  {
 
   void shlxl(Register dst, Register src1, Register src2);
   void shlxq(Register dst, Register src1, Register src2);
+  void shrxl(Register dst, Register src1, Register src2);
   void shrxq(Register dst, Register src1, Register src2);
 
   void bzhiq(Register dst, Register src1, Register src2);