8241597: x86: Remove MMX support

Reviewed-by: kvn, redestad

Author: Vladimir Ivanov

src/hotspot/cpu/x86/assembler_x86.cpp

@@ -1163,19 +1163,6 @@ void Assembler::emit_operand(XMMRegister reg, Address adr) {
     }
 }
 
-// MMX operations
-void Assembler::emit_operand(MMXRegister reg, Address adr) {
-  assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
-  emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
-}
-
-// work around gcc (3.2.1-7a) bug
-void Assembler::emit_operand(Address adr, MMXRegister reg) {
-  assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers");
-  emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec);
-}
-
-
 // Now the Assembler instructions (identical for 32/64 bits)
 
 void Assembler::adcl(Address dst, int32_t imm32) {
@@ -2003,11 +1990,6 @@ void Assembler::divss(XMMRegister dst, XMMRegister src) {
   emit_int16(0x5E, (0xC0 | encode));
 }
 
-void Assembler::emms() {
-  NOT_LP64(assert(VM_Version::supports_mmx(), ""));
-  emit_int16(0x0F, 0x77);
-}
-
 void Assembler::hlt() {
   emit_int8((unsigned char)0xF4);
 }
@@ -2771,24 +2753,6 @@ void Assembler::movlpd(XMMRegister dst, Address src) {
   emit_operand(dst, src);
 }
 
-void Assembler::movq( MMXRegister dst, Address src ) {
-  assert( VM_Version::supports_mmx(), "" );
-  emit_int16(0x0F, 0x6F);
-  emit_operand(dst, src);
-}
-
-void Assembler::movq( Address dst, MMXRegister src ) {
-  assert( VM_Version::supports_mmx(), "" );
-  emit_int16(0x0F, 0x7F);
-  // workaround gcc (3.2.1-7a) bug
-  // In that version of gcc with only an emit_operand(MMX, Address)
-  // gcc will tail jump and try and reverse the parameters completely
-  // obliterating dst in the process. By having a version available
-  // that doesn't need to swap the args at the tail jump the bug is
-  // avoided.
-  emit_operand(dst, src);
-}
-
 void Assembler::movq(XMMRegister dst, Address src) {
   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
   InstructionMark im(this);
@@ -6861,6 +6825,11 @@ void Assembler::vzeroupper_uncached() {
 #ifndef _LP64
 // 32bit only pieces of the assembler
 
+void Assembler::emms() {
+  NOT_LP64(assert(VM_Version::supports_mmx(), ""));
+  emit_int16(0x0F, 0x77);
+}
+
 void Assembler::vzeroupper() {
   vzeroupper_uncached();
 }

src/hotspot/cpu/x86/assembler_x86.hpp

@@ -766,11 +766,6 @@ class Assembler : public AbstractAssembler  {
 
   void emit_operand(XMMRegister reg, Address adr);
 
-  void emit_operand(MMXRegister reg, Address adr);
-
-  // workaround gcc (3.2.1-7) bug
-  void emit_operand(Address adr, MMXRegister reg);
-
   // Immediate-to-memory forms
   void emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32);
 
@@ -1168,7 +1163,6 @@ class Assembler : public AbstractAssembler  {
   void divss(XMMRegister dst, Address src);
   void divss(XMMRegister dst, XMMRegister src);
 
-  void emms();
 
 #ifndef _LP64
  private:
@@ -1178,6 +1172,8 @@ class Assembler : public AbstractAssembler  {
   void emit_farith(int b1, int b2, int i);
 
  public:
+  void emms();
+
   void fabs();
 
   void fadd(int i);
@@ -1543,12 +1539,7 @@ class Assembler : public AbstractAssembler  {
   void movq(Register dst, Register src);
   void movq(Register dst, Address src);
   void movq(Address  dst, Register src);
-#endif
 
-  void movq(Address     dst, MMXRegister src );
-  void movq(MMXRegister dst, Address src );
-
-#ifdef _LP64
   // These dummies prevent using movq from converting a zero (like NULL) into Register
   // by giving the compiler two choices it can't resolve
 

src/hotspot/cpu/x86/register_definitions_x86.cpp

@@ -129,16 +129,6 @@ REGISTER_DEFINITION(Register, r12_heapbase);
 REGISTER_DEFINITION(Register, r15_thread);
 #endif // AMD64
 
-REGISTER_DEFINITION(MMXRegister, mnoreg );
-REGISTER_DEFINITION(MMXRegister, mmx0 );
-REGISTER_DEFINITION(MMXRegister, mmx1 );
-REGISTER_DEFINITION(MMXRegister, mmx2 );
-REGISTER_DEFINITION(MMXRegister, mmx3 );
-REGISTER_DEFINITION(MMXRegister, mmx4 );
-REGISTER_DEFINITION(MMXRegister, mmx5 );
-REGISTER_DEFINITION(MMXRegister, mmx6 );
-REGISTER_DEFINITION(MMXRegister, mmx7 );
-
 REGISTER_DEFINITION(KRegister, knoreg);
 REGISTER_DEFINITION(KRegister, k0);
 REGISTER_DEFINITION(KRegister, k1);

src/hotspot/cpu/x86/register_x86.hpp

@@ -130,17 +130,10 @@ CONSTANT_REGISTER_DECLARATION(FloatRegister, fnoreg, (-1));
 class XMMRegisterImpl;
 typedef XMMRegisterImpl* XMMRegister;
 
-// Use MMXRegister as shortcut
-class MMXRegisterImpl;
-typedef MMXRegisterImpl* MMXRegister;
-
 inline XMMRegister as_XMMRegister(int encoding) {
   return (XMMRegister)(intptr_t)encoding;
 }
 
-inline MMXRegister as_MMXRegister(int encoding) {
-  return (MMXRegister)(intptr_t)encoding;
-}
 
 // The implementation of XMM registers for the IA32 architecture
 class XMMRegisterImpl: public AbstractRegisterImpl {
@@ -212,17 +205,6 @@ CONSTANT_REGISTER_DECLARATION(XMMRegister, xmm31,    (31));
 // can't be described in oopMaps and therefore can't be used by the compilers (at least
 // were deopt might wan't to see them).
 
-// The MMX registers, for P3 and up chips
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mnoreg , (-1));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx0 , ( 0));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx1 , ( 1));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx2 , ( 2));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx3 , ( 3));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx4 , ( 4));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx5 , ( 5));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx6 , ( 6));
-CONSTANT_REGISTER_DECLARATION(MMXRegister, mmx7 , ( 7));
-
 // Use XMMRegister as shortcut
 class KRegisterImpl;
 typedef KRegisterImpl* KRegister;

src/hotspot/cpu/x86/stubGenerator_x86_32.cpp

@@ -795,55 +795,6 @@ class StubGenerator: public StubCodeGenerator {
   __ BIND(L_exit);
   }
 
-  // Copy 64 bytes chunks
-  //
-  // Inputs:
-  //   from        - source array address
-  //   to_from     - destination array address - from
-  //   qword_count - 8-bytes element count, negative
-  //
-  void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
-    assert( VM_Version::supports_mmx(), "supported cpu only" );
-    Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
-    // Copy 64-byte chunks
-    __ jmpb(L_copy_64_bytes);
-    __ align(OptoLoopAlignment);
-  __ BIND(L_copy_64_bytes_loop);
-    __ movq(mmx0, Address(from, 0));
-    __ movq(mmx1, Address(from, 8));
-    __ movq(mmx2, Address(from, 16));
-    __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
-    __ movq(mmx3, Address(from, 24));
-    __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
-    __ movq(mmx4, Address(from, 32));
-    __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
-    __ movq(mmx5, Address(from, 40));
-    __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
-    __ movq(mmx6, Address(from, 48));
-    __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
-    __ movq(mmx7, Address(from, 56));
-    __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
-    __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
-    __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
-    __ addptr(from, 64);
-  __ BIND(L_copy_64_bytes);
-    __ subl(qword_count, 8);
-    __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
-    __ addl(qword_count, 8);
-    __ jccb(Assembler::zero, L_exit);
-    //
-    // length is too short, just copy qwords
-    //
-  __ BIND(L_copy_8_bytes);
-    __ movq(mmx0, Address(from, 0));
-    __ movq(Address(from, to_from, Address::times_1), mmx0);
-    __ addptr(from, 8);
-    __ decrement(qword_count);
-    __ jcc(Assembler::greater, L_copy_8_bytes);
-  __ BIND(L_exit);
-    __ emms();
-  }
-
   address generate_disjoint_copy(BasicType t, bool aligned,
                                  Address::ScaleFactor sf,
                                  address* entry, const char *name,
@@ -918,7 +869,7 @@ class StubGenerator: public StubCodeGenerator {
         __ subl(count, 1<<(shift-1));
       __ BIND(L_skip_align2);
       }
-      if (!VM_Version::supports_mmx()) {
+      if (!UseXMMForArrayCopy) {
         __ mov(rax, count);      // save 'count'
         __ shrl(count, shift); // bytes count
         __ addptr(to_from, from);// restore 'to'
@@ -935,18 +886,14 @@ class StubGenerator: public StubCodeGenerator {
           __ movl(Address(from, to_from, Address::times_1, 0), rax);
           __ addptr(from, 4);
           __ subl(count, 1<<shift);
-         }
+        }
       __ BIND(L_copy_64_bytes);
         __ mov(rax, count);
         __ shrl(rax, shift+1);  // 8 bytes chunk count
         //
-        // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
+        // Copy 8-byte chunks through XMM registers, 8 per iteration of the loop
         //
-        if (UseXMMForArrayCopy) {
-          xmm_copy_forward(from, to_from, rax);
-        } else {
-          mmx_copy_forward(from, to_from, rax);
-        }
+        xmm_copy_forward(from, to_from, rax);
       }
       // copy tailing dword
     __ BIND(L_copy_4_bytes);
@@ -979,9 +926,6 @@ class StubGenerator: public StubCodeGenerator {
       }
     }
 
-    if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
-      __ emms();
-    }
     __ movl(count, Address(rsp, 12+12)); // reread 'count'
     bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
 
@@ -1117,7 +1061,7 @@ class StubGenerator: public StubCodeGenerator {
         __ jcc(Assembler::below, L_copy_4_bytes);
       }
 
-      if (!VM_Version::supports_mmx()) {
+      if (!UseXMMForArrayCopy) {
         __ std();
         __ mov(rax, count); // Save 'count'
         __ mov(rdx, to);    // Save 'to'
@@ -1143,20 +1087,12 @@ class StubGenerator: public StubCodeGenerator {
         __ align(OptoLoopAlignment);
         // Move 8 bytes
       __ BIND(L_copy_8_bytes_loop);
-        if (UseXMMForArrayCopy) {
-          __ movq(xmm0, Address(from, count, sf, 0));
-          __ movq(Address(to, count, sf, 0), xmm0);
-        } else {
-          __ movq(mmx0, Address(from, count, sf, 0));
-          __ movq(Address(to, count, sf, 0), mmx0);
-        }
+        __ movq(xmm0, Address(from, count, sf, 0));
+        __ movq(Address(to, count, sf, 0), xmm0);
       __ BIND(L_copy_8_bytes);
         __ subl(count, 2<<shift);
         __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
         __ addl(count, 2<<shift);
-        if (!UseXMMForArrayCopy) {
-          __ emms();
-        }
       }
     __ BIND(L_copy_4_bytes);
       // copy prefix qword
@@ -1190,9 +1126,6 @@ class StubGenerator: public StubCodeGenerator {
       }
     }
 
-    if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
-      __ emms();
-    }
     __ movl2ptr(count, Address(rsp, 12+12)); // reread count
     bs->arraycopy_epilogue(_masm, decorators, t, from, to, count);
 
@@ -1232,12 +1165,8 @@ class StubGenerator: public StubCodeGenerator {
       // UnsafeCopyMemory page error: continue after ucm
       UnsafeCopyMemoryMark ucmm(this, true, true);
       __ subptr(to, from); // to --> to_from
-      if (VM_Version::supports_mmx()) {
-        if (UseXMMForArrayCopy) {
-          xmm_copy_forward(from, to_from, count);
-        } else {
-          mmx_copy_forward(from, to_from, count);
-        }
+      if (UseXMMForArrayCopy) {
+        xmm_copy_forward(from, to_from, count);
       } else {
         __ jmpb(L_copy_8_bytes);
         __ align(OptoLoopAlignment);
@@ -1250,9 +1179,6 @@ class StubGenerator: public StubCodeGenerator {
         __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
       }
     }
-    if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
-      __ emms();
-    }
     inc_copy_counter_np(T_LONG);
     __ leave(); // required for proper stackwalking of RuntimeStub frame
     __ vzeroupper();
@@ -1298,14 +1224,9 @@ class StubGenerator: public StubCodeGenerator {
 
       __ align(OptoLoopAlignment);
     __ BIND(L_copy_8_bytes_loop);
-      if (VM_Version::supports_mmx()) {
-        if (UseXMMForArrayCopy) {
-          __ movq(xmm0, Address(from, count, Address::times_8));
-          __ movq(Address(to, count, Address::times_8), xmm0);
-        } else {
-          __ movq(mmx0, Address(from, count, Address::times_8));
-          __ movq(Address(to, count, Address::times_8), mmx0);
-        }
+      if (UseXMMForArrayCopy) {
+        __ movq(xmm0, Address(from, count, Address::times_8));
+        __ movq(Address(to, count, Address::times_8), xmm0);
       } else {
         __ fild_d(Address(from, count, Address::times_8));
         __ fistp_d(Address(to, count, Address::times_8));
@@ -1315,9 +1236,6 @@ class StubGenerator: public StubCodeGenerator {
       __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
 
     }
-    if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
-      __ emms();
-    }
     inc_copy_counter_np(T_LONG);
     __ leave(); // required for proper stackwalking of RuntimeStub frame
     __ xorptr(rax, rax); // return 0

src/hotspot/cpu/x86/vm_version_x86.cpp

@@ -734,7 +734,7 @@ void VM_Version::get_processor_features() {
   char buf[512];
   int res = jio_snprintf(buf, sizeof(buf),
               "(%u cores per cpu, %u threads per core) family %d model %d stepping %d"
-              "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s",
+              "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s%s%s%s%s" "%s%s%s%s%s%s",
 
                cores_per_cpu(), threads_per_core(),
                cpu_family(), _model, _stepping,
@@ -758,10 +758,9 @@ void VM_Version::get_processor_features() {
                (supports_clmul()  ? ", clmul" : ""),
                (supports_erms()   ? ", erms" : ""),
                (supports_rtm()    ? ", rtm" : ""),
-               (supports_mmx_ext() ? ", mmxext" : ""),
                (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
-
                (supports_lzcnt()   ? ", lzcnt": ""),
+
                (supports_sse4a()   ? ", sse4a": ""),
                (supports_ht() ? ", ht": ""),
                (supports_tsc() ? ", tsc": ""),
@@ -771,8 +770,8 @@ void VM_Version::get_processor_features() {
                (supports_bmi2() ? ", bmi2" : ""),
                (supports_adx() ? ", adx" : ""),
                (supports_evex() ? ", avx512f" : ""),
-
                (supports_avx512dq() ? ", avx512dq" : ""),
+
                (supports_avx512pf() ? ", avx512pf" : ""),
                (supports_avx512er() ? ", avx512er" : ""),
                (supports_avx512cd() ? ", avx512cd" : ""),
@@ -782,8 +781,8 @@ void VM_Version::get_processor_features() {
                (supports_avx512_vpclmulqdq() ? ", avx512_vpclmulqdq" : ""),
                (supports_avx512_vbmi() ? ", avx512_vbmi" : ""),
                (supports_avx512_vbmi2() ? ", avx512_vbmi2" : ""),
-
                (supports_avx512_vaes() ? ", avx512_vaes" : ""),
+
                (supports_avx512_vnni() ? ", avx512_vnni" : ""),
                (supports_sha() ? ", sha" : ""),
                (supports_fma() ? ", fma" : ""),

src/hotspot/cpu/x86/vm_version_x86.hpp

@@ -905,7 +905,6 @@ enum Extended_Family {
 
   // AMD features
   static bool supports_3dnow_prefetch()    { return (_features & CPU_3DNOW_PREFETCH) != 0; }
-  static bool supports_mmx_ext()  { return is_amd_family() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
   static bool supports_lzcnt()    { return (_features & CPU_LZCNT) != 0; }
   static bool supports_sse4a()    { return (_features & CPU_SSE4A) != 0; }