6940701: Don't align loops in stubs for Niagara sparc

Summary: Don't align loops in stubs for Niagara sparc since NOPs are expensive.
Reviewed-by: twisti, never

src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp

@@ -2849,7 +2849,7 @@ void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
 
 
 void LIR_Assembler::align_backward_branch_target() {
-  __ align(16);
+  __ align(OptoLoopAlignment);
 }
 
 

src/cpu/sparc/vm/c2_globals_sparc.hpp

@@ -60,9 +60,6 @@ define_pd_global(intx, FreqInlineSize,               175);
 define_pd_global(intx, INTPRESSURE,                  48);  // large register set
 define_pd_global(intx, InteriorEntryAlignment,       16);  // = CodeEntryAlignment
 define_pd_global(intx, NewSizeThreadIncrease, ScaleForWordSize(4*K));
-// The default setting 16/16 seems to work best.
-// (For _228_jack 16/16 is 2% better than 4/4, 16/4, 32/32, 32/16, or 16/32.)
-define_pd_global(intx, OptoLoopAlignment,            16);  // = 4*wordSize
 define_pd_global(intx, RegisterCostAreaRatio,        12000);
 define_pd_global(bool, UseTLAB,                      true);
 define_pd_global(bool, ResizeTLAB,                   true);

src/cpu/sparc/vm/globals_sparc.hpp

@@ -40,6 +40,9 @@ define_pd_global(bool, ImplicitNullChecks,          true);  // Generate code for
 define_pd_global(bool, UncommonNullCast,            true);  // Uncommon-trap NULLs past to check cast
 
 define_pd_global(intx, CodeEntryAlignment,    32);
+// The default setting 16/16 seems to work best.
+// (For _228_jack 16/16 is 2% better than 4/4, 16/4, 32/32, 32/16, or 16/32.)
+define_pd_global(intx, OptoLoopAlignment,     16);  // = 4*wordSize
 define_pd_global(intx, InlineFrequencyCount,  50);  // we can use more inlining on the SPARC
 define_pd_global(intx, InlineSmallCode,       1500);
 #ifdef _LP64

src/cpu/sparc/vm/stubGenerator_sparc.cpp

@@ -1148,7 +1148,7 @@ class StubGenerator: public StubCodeGenerator {
       __ andn(from, 7, from);     // Align address
       __ ldx(from, 0, O3);
       __ inc(from, 8);
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_loop);
       __ ldx(from, 0, O4);
       __ deccc(count, count_dec); // Can we do next iteration after this one?
@@ -1220,7 +1220,7 @@ class StubGenerator: public StubCodeGenerator {
     //
       __ andn(end_from, 7, end_from);     // Align address
       __ ldx(end_from, 0, O3);
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_loop);
       __ ldx(end_from, -8, O4);
       __ deccc(count, count_dec); // Can we do next iteration after this one?
@@ -1349,7 +1349,7 @@ class StubGenerator: public StubCodeGenerator {
     __ BIND(L_copy_byte);
       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
       __ delayed()->nop();
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_byte_loop);
       __ ldub(from, offset, O3);
       __ deccc(count);
@@ -1445,7 +1445,7 @@ class StubGenerator: public StubCodeGenerator {
                                         L_aligned_copy, L_copy_byte);
     }
     // copy 4 elements (16 bytes) at a time
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_aligned_copy);
       __ dec(end_from, 16);
       __ ldx(end_from, 8, O3);
@@ -1461,7 +1461,7 @@ class StubGenerator: public StubCodeGenerator {
     __ BIND(L_copy_byte);
       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
       __ delayed()->nop();
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_byte_loop);
       __ dec(end_from);
       __ dec(end_to);
@@ -1577,7 +1577,7 @@ class StubGenerator: public StubCodeGenerator {
     __ BIND(L_copy_2_bytes);
       __ br_zero(Assembler::zero, false, Assembler::pt, count, L_exit);
       __ delayed()->nop();
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_2_bytes_loop);
       __ lduh(from, offset, O3);
       __ deccc(count);
@@ -1684,7 +1684,7 @@ class StubGenerator: public StubCodeGenerator {
                                         L_aligned_copy, L_copy_2_bytes);
     }
     // copy 4 elements (16 bytes) at a time
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_aligned_copy);
       __ dec(end_from, 16);
       __ ldx(end_from, 8, O3);
@@ -1781,7 +1781,7 @@ class StubGenerator: public StubCodeGenerator {
     // copy with shift 4 elements (16 bytes) at a time
       __ dec(count, 4);   // The cmp at the beginning guaranty count >= 4
 
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_16_bytes);
       __ ldx(from, 4, O4);
       __ deccc(count, 4); // Can we do next iteration after this one?
@@ -1907,7 +1907,7 @@ class StubGenerator: public StubCodeGenerator {
     // to form 2 aligned 8-bytes chunks to store.
     //
       __ ldx(end_from, -4, O3);
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_16_bytes);
       __ ldx(end_from, -12, O4);
       __ deccc(count, 4);
@@ -1929,7 +1929,7 @@ class StubGenerator: public StubCodeGenerator {
       __ delayed()->inc(count, 4);
 
     // copy 4 elements (16 bytes) at a time
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_aligned_copy);
       __ dec(end_from, 16);
       __ ldx(end_from, 8, O3);
@@ -2045,7 +2045,7 @@ class StubGenerator: public StubCodeGenerator {
       __ mov(O3, count);
       __ mov(from, from64);
 
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_64_bytes);
       for( int off = 0; off < 64; off += 16 ) {
         __ ldx(from64,  off+0, O4);
@@ -2065,7 +2065,7 @@ class StubGenerator: public StubCodeGenerator {
       __ delayed()->add(offset0, 8, offset8);
 
       // Copy by 16 bytes chunks
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_16_bytes);
       __ ldx(from, offset0, O3);
       __ ldx(from, offset8, G3);
@@ -2139,7 +2139,7 @@ class StubGenerator: public StubCodeGenerator {
       __ brx(Assembler::lessEqual, false, Assembler::pn, L_copy_8_bytes );
       __ delayed()->sllx(count, LogBytesPerLong, offset8);
       __ sub(offset8, 8, offset0);
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_16_bytes);
       __ ldx(from, offset8, O2);
       __ ldx(from, offset0, O3);
@@ -2405,7 +2405,7 @@ class StubGenerator: public StubCodeGenerator {
     //   (O5 = 0; ; O5 += wordSize) --- offset from src, dest arrays
     //   (O2 = len; O2 != 0; O2--) --- number of oops *remaining*
     //   G3, G4, G5 --- current oop, oop.klass, oop.klass.super
-    __ align(16);
+    __ align(OptoLoopAlignment);
 
     __ BIND(store_element);
     __ deccc(G1_remain);                // decrement the count

src/cpu/sparc/vm/vm_version_sparc.cpp

@@ -86,14 +86,14 @@ void VM_Version::initialize() {
     if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) {
       FLAG_SET_DEFAULT(InteriorEntryAlignment, 4);
     }
-    if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
-      FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
-    }
     if (is_niagara1_plus() && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
       // Use smaller prefetch distance on N2
       FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
     }
 #endif
+    if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
+      FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
+    }
   }
 
   // Use hardware population count instruction if available.

src/cpu/x86/vm/c2_globals_x86.hpp

@@ -80,7 +80,6 @@ define_pd_global(intx, CodeCacheExpansionSize,       32*K);
 // Ergonomics related flags
 define_pd_global(uint64_t,MaxRAM,                    4ULL*G);
 #endif // AMD64
-define_pd_global(intx, OptoLoopAlignment,            16);
 define_pd_global(intx, RegisterCostAreaRatio,        16000);
 
 // Peephole and CISC spilling both break the graph, and so makes the

src/cpu/x86/vm/globals_x86.hpp

@@ -45,6 +45,7 @@ define_pd_global(intx, CodeEntryAlignment,       32);
 #else
 define_pd_global(intx, CodeEntryAlignment,       16);
 #endif // COMPILER2
+define_pd_global(intx, OptoLoopAlignment,        16);
 define_pd_global(intx, InlineFrequencyCount,     100);
 define_pd_global(intx, InlineSmallCode,          1000);
 

src/cpu/x86/vm/stubGenerator_x86_32.cpp

@@ -812,7 +812,7 @@ class StubGenerator: public StubCodeGenerator {
     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(16);
+    __ align(OptoLoopAlignment);
   __ BIND(L_copy_64_bytes_loop);
 
     if(UseUnalignedLoadStores) {
@@ -874,7 +874,7 @@ class StubGenerator: public StubCodeGenerator {
     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(16);
+    __ align(OptoLoopAlignment);
   __ BIND(L_copy_64_bytes_loop);
     __ movq(mmx0, Address(from, 0));
     __ movq(mmx1, Address(from, 8));
@@ -1144,7 +1144,7 @@ class StubGenerator: public StubCodeGenerator {
       __ movl(Address(to, count, sf, 0), rdx);
       __ jmpb(L_copy_8_bytes);
 
-      __ align(16);
+      __ align(OptoLoopAlignment);
       // Move 8 bytes
     __ BIND(L_copy_8_bytes_loop);
       if (UseXMMForArrayCopy) {
@@ -1235,7 +1235,7 @@ class StubGenerator: public StubCodeGenerator {
       }
     } else {
       __ jmpb(L_copy_8_bytes);
-      __ align(16);
+      __ align(OptoLoopAlignment);
     __ BIND(L_copy_8_bytes_loop);
       __ fild_d(Address(from, 0));
       __ fistp_d(Address(from, to_from, Address::times_1));
@@ -1282,7 +1282,7 @@ class StubGenerator: public StubCodeGenerator {
 
     __ jmpb(L_copy_8_bytes);
 
-    __ align(16);
+    __ align(OptoLoopAlignment);
   __ BIND(L_copy_8_bytes_loop);
     if (VM_Version::supports_mmx()) {
       if (UseXMMForArrayCopy) {
@@ -1454,7 +1454,7 @@ class StubGenerator: public StubCodeGenerator {
     // Loop control:
     //   for (count = -count; count != 0; count++)
     // Base pointers src, dst are biased by 8*count,to last element.
-    __ align(16);
+    __ align(OptoLoopAlignment);
 
     __ BIND(L_store_element);
     __ movptr(to_element_addr, elem);     // store the oop

src/cpu/x86/vm/stubGenerator_x86_64.cpp

@@ -871,9 +871,8 @@ class StubGenerator: public StubCodeGenerator {
   }
 
   address generate_fp_mask(const char *stub_name, int64_t mask) {
+    __ align(CodeEntryAlignment);
     StubCodeMark mark(this, "StubRoutines", stub_name);
-
-    __ align(16);
     address start = __ pc();
 
     __ emit_data64( mask, relocInfo::none );
@@ -1268,7 +1267,7 @@ class StubGenerator: public StubCodeGenerator {
                              Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
     DEBUG_ONLY(__ stop("enter at entry label, not here"));
     Label L_loop;
-    __ align(16);
+    __ align(OptoLoopAlignment);
   __ BIND(L_loop);
     if(UseUnalignedLoadStores) {
       __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
@@ -1309,7 +1308,7 @@ class StubGenerator: public StubCodeGenerator {
                               Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
     DEBUG_ONLY(__ stop("enter at entry label, not here"));
     Label L_loop;
-    __ align(16);
+    __ align(OptoLoopAlignment);
   __ BIND(L_loop);
     if(UseUnalignedLoadStores) {
       __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
@@ -2229,7 +2228,7 @@ class StubGenerator: public StubCodeGenerator {
     // Loop control:
     //   for (count = -count; count != 0; count++)
     // Base pointers src, dst are biased by 8*(count-1),to last element.
-    __ align(16);
+    __ align(OptoLoopAlignment);
 
     __ BIND(L_store_element);
     __ store_heap_oop(to_element_addr, rax_oop);  // store the oop

src/share/vm/opto/c2_globals.hpp

@@ -52,9 +52,6 @@
           "Code alignment for interior entry points "                       \
           "in generated code (in bytes)")                                   \
                                                                             \
-  product_pd(intx, OptoLoopAlignment,                                       \
-          "Align inner loops to zero relative to this modulus")             \
-                                                                            \
   product(intx, MaxLoopPad, (OptoLoopAlignment-1),                          \
           "Align a loop if padding size in bytes is less or equal to this value") \
                                                                             \

src/share/vm/runtime/globals.hpp

@@ -3110,6 +3110,9 @@ class CommandLineFlags {
   develop_pd(intx, CodeEntryAlignment,                                      \
           "Code entry alignment for generated code (in bytes)")             \
                                                                             \
+  product_pd(intx, OptoLoopAlignment,                                       \
+          "Align inner loops to zero relative to this modulus")             \
+                                                                            \
   product_pd(uintx, InitialCodeCacheSize,                                   \
           "Initial code cache size (in bytes)")                             \
                                                                             \