--- a/hotspot/src/cpu/ppc/vm/c1_MacroAssembler_ppc.cpp Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/cpu/ppc/vm/c1_MacroAssembler_ppc.cpp Mon Dec 19 00:49:34 2016 +0100
@@ -238,72 +238,15 @@
int obj_size_in_bytes, int hdr_size_in_bytes) {
const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
- const int cl_size = VM_Version::L1_data_cache_line_size(),
- cl_dwords = cl_size>>3,
- cl_dw_addr_bits = exact_log2(cl_dwords);
-
- const Register tmp = R0,
- base_ptr = tmp1,
- cnt_dwords = tmp2;
-
- if (index <= 6) {
- // Use explicit NULL stores.
- if (index > 0) { li(tmp, 0); }
- for (int i = 0; i < index; ++i) { std(tmp, hdr_size_in_bytes + i * HeapWordSize, obj); }
-
- } else if (index < (2<<cl_dw_addr_bits)-1) {
- // simple loop
- Label loop;
-
- li(cnt_dwords, index);
- addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
- li(tmp, 0);
- mtctr(cnt_dwords); // Load counter.
- bind(loop);
- std(tmp, 0, base_ptr); // Clear 8byte aligned block.
- addi(base_ptr, base_ptr, 8);
- bdnz(loop);
-
+ // 2x unrolled loop is shorter with more than 9 HeapWords.
+ if (index <= 9) {
+ clear_memory_unrolled(obj, index, R0, hdr_size_in_bytes);
} else {
- // like clear_memory_doubleword
- Label startloop, fast, fastloop, restloop, done;
+ const Register base_ptr = tmp1,
+ cnt_dwords = tmp2;
- addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
- load_const_optimized(cnt_dwords, index);
- rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
- beq(CCR0, fast); // Already 128byte aligned.
-
- subfic(tmp, tmp, cl_dwords);
- mtctr(tmp); // Set ctr to hit 128byte boundary (0<ctr<cl_dwords).
- subf(cnt_dwords, tmp, cnt_dwords); // rest.
- li(tmp, 0);
-
- bind(startloop); // Clear at the beginning to reach 128byte boundary.
- std(tmp, 0, base_ptr); // Clear 8byte aligned block.
- addi(base_ptr, base_ptr, 8);
- bdnz(startloop);
-
- bind(fast); // Clear 128byte blocks.
- srdi(tmp, cnt_dwords, cl_dw_addr_bits); // Loop count for 128byte loop (>0).
- andi(cnt_dwords, cnt_dwords, cl_dwords-1); // Rest in dwords.
- mtctr(tmp); // Load counter.
-
- bind(fastloop);
- dcbz(base_ptr); // Clear 128byte aligned block.
- addi(base_ptr, base_ptr, cl_size);
- bdnz(fastloop);
-
- cmpdi(CCR0, cnt_dwords, 0); // size 0?
- beq(CCR0, done); // rest == 0
- li(tmp, 0);
- mtctr(cnt_dwords); // Load counter.
-
- bind(restloop); // Clear rest.
- std(tmp, 0, base_ptr); // Clear 8byte aligned block.
- addi(base_ptr, base_ptr, 8);
- bdnz(restloop);
-
- bind(done);
+ addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
+ clear_memory_doubleword(base_ptr, cnt_dwords, R0, index);
}
}
--- a/hotspot/src/cpu/ppc/vm/globals_ppc.hpp Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/cpu/ppc/vm/globals_ppc.hpp Mon Dec 19 00:49:34 2016 +0100
@@ -77,7 +77,8 @@
define_pd_global(bool, CompactStrings, true);
-define_pd_global(intx, InitArrayShortSize, 8*BytesPerLong);
+// 2x unrolled loop is shorter with more than 9 HeapWords.
+define_pd_global(intx, InitArrayShortSize, 9*BytesPerLong);
// Platform dependent flag handling: flags only defined on this platform.
#define ARCH_FLAGS(develop, \
--- a/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.cpp Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.cpp Mon Dec 19 00:49:34 2016 +0100
@@ -3332,53 +3332,90 @@
}
// Clear Array
+// For very short arrays. tmp == R0 is allowed.
+void MacroAssembler::clear_memory_unrolled(Register base_ptr, int cnt_dwords, Register tmp, int offset) {
+ if (cnt_dwords > 0) { li(tmp, 0); }
+ for (int i = 0; i < cnt_dwords; ++i) { std(tmp, offset + i * 8, base_ptr); }
+}
+
+// Version for constant short array length. Kills base_ptr. tmp == R0 is allowed.
+void MacroAssembler::clear_memory_constlen(Register base_ptr, int cnt_dwords, Register tmp) {
+ if (cnt_dwords < 8) {
+ clear_memory_unrolled(base_ptr, cnt_dwords, tmp);
+ return;
+ }
+
+ Label loop;
+ const long loopcnt = cnt_dwords >> 1,
+ remainder = cnt_dwords & 1;
+
+ li(tmp, loopcnt);
+ mtctr(tmp);
+ li(tmp, 0);
+ bind(loop);
+ std(tmp, 0, base_ptr);
+ std(tmp, 8, base_ptr);
+ addi(base_ptr, base_ptr, 16);
+ bdnz(loop);
+ if (remainder) { std(tmp, 0, base_ptr); }
+}
+
// Kills both input registers. tmp == R0 is allowed.
-void MacroAssembler::clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp) {
+void MacroAssembler::clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp, long const_cnt) {
// Procedure for large arrays (uses data cache block zero instruction).
Label startloop, fast, fastloop, small_rest, restloop, done;
const int cl_size = VM_Version::L1_data_cache_line_size(),
- cl_dwords = cl_size>>3,
+ cl_dwords = cl_size >> 3,
cl_dw_addr_bits = exact_log2(cl_dwords),
- dcbz_min = 1; // Min count of dcbz executions, needs to be >0.
-
-//2:
- cmpdi(CCR1, cnt_dwords, ((dcbz_min+1)<<cl_dw_addr_bits)-1); // Big enough? (ensure >=dcbz_min lines included).
- blt(CCR1, small_rest); // Too small.
- rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
- beq(CCR0, fast); // Already 128byte aligned.
+ dcbz_min = 1, // Min count of dcbz executions, needs to be >0.
+ min_cnt = ((dcbz_min + 1) << cl_dw_addr_bits) - 1;
+
+ if (const_cnt >= 0) {
+ // Constant case.
+ if (const_cnt < min_cnt) {
+ clear_memory_constlen(base_ptr, const_cnt, tmp);
+ return;
+ }
+ load_const_optimized(cnt_dwords, const_cnt, tmp);
+ } else {
+ // cnt_dwords already loaded in register. Need to check size.
+ cmpdi(CCR1, cnt_dwords, min_cnt); // Big enough? (ensure >= dcbz_min lines included).
+ blt(CCR1, small_rest);
+ }
+ rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
+ beq(CCR0, fast); // Already 128byte aligned.
subfic(tmp, tmp, cl_dwords);
mtctr(tmp); // Set ctr to hit 128byte boundary (0<ctr<cl_dwords).
subf(cnt_dwords, tmp, cnt_dwords); // rest.
li(tmp, 0);
-//10:
+
bind(startloop); // Clear at the beginning to reach 128byte boundary.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(startloop);
-//13:
+
bind(fast); // Clear 128byte blocks.
srdi(tmp, cnt_dwords, cl_dw_addr_bits); // Loop count for 128byte loop (>0).
andi(cnt_dwords, cnt_dwords, cl_dwords-1); // Rest in dwords.
mtctr(tmp); // Load counter.
-//16:
+
bind(fastloop);
dcbz(base_ptr); // Clear 128byte aligned block.
addi(base_ptr, base_ptr, cl_size);
bdnz(fastloop);
- if (InsertEndGroupPPC64) { endgroup(); } else { nop(); }
-//20:
+
bind(small_rest);
cmpdi(CCR0, cnt_dwords, 0); // size 0?
beq(CCR0, done); // rest == 0
li(tmp, 0);
mtctr(cnt_dwords); // Load counter.
-//24:
+
bind(restloop); // Clear rest.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(restloop);
-//27:
+
bind(done);
}
--- a/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.hpp Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.hpp Mon Dec 19 00:49:34 2016 +0100
@@ -755,7 +755,9 @@
is_trap_range_check_g(x) || is_trap_range_check_ge(x);
}
- void clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp = R0);
+ void clear_memory_unrolled(Register base_ptr, int cnt_dwords, Register tmp = R0, int offset = 0);
+ void clear_memory_constlen(Register base_ptr, int cnt_dwords, Register tmp = R0);
+ void clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp = R0, long const_cnt = -1);
#ifdef COMPILER2
// Intrinsics for CompactStrings
--- a/hotspot/src/cpu/ppc/vm/ppc.ad Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/cpu/ppc/vm/ppc.ad Mon Dec 19 00:49:34 2016 +0100
@@ -965,10 +965,7 @@
// is the number of bytes (not instructions) which will be inserted before
// the instruction. The padding must match the size of a NOP instruction.
-int inlineCallClearArrayNode::compute_padding(int current_offset) const {
- int desired_padding = (2*4-current_offset)&31; // see MacroAssembler::clear_memory_doubleword
- return (desired_padding <= 3*4) ? desired_padding : 0;
-}
+// Currently not used on this platform.
//=============================================================================
@@ -4066,6 +4063,14 @@
interface(CONST_INTER);
%}
+operand immLmax30() %{
+ predicate((n->get_long() <= 30));
+ match(ConL);
+ op_cost(0);
+ format %{ %}
+ interface(CONST_INTER);
+%}
+
// Long Immediate: 16-bit
operand immL16() %{
predicate(Assembler::is_simm(n->get_long(), 16));
@@ -11735,18 +11740,44 @@
ins_pipe(pipe_class_default);
%}
-// Clear-array with dynamic array-size.
+// Clear-array with constant short array length. The versions below can use dcbz with cnt > 30.
+instruct inlineCallClearArrayShort(immLmax30 cnt, rarg2RegP base, Universe dummy, regCTR ctr) %{
+ match(Set dummy (ClearArray cnt base));
+ effect(USE_KILL base, KILL ctr);
+ ins_cost(2 * MEMORY_REF_COST);
+
+ format %{ "ClearArray $cnt, $base" %}
+ ins_encode %{
+ // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+ __ clear_memory_constlen($base$$Register, $cnt$$constant, R0); // kills base, R0
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+// Clear-array with constant large array length.
+instruct inlineCallClearArrayLarge(immL cnt, rarg2RegP base, Universe dummy, iRegLdst tmp, regCTR ctr) %{
+ match(Set dummy (ClearArray cnt base));
+ effect(USE_KILL base, TEMP tmp, KILL ctr);
+ ins_cost(3 * MEMORY_REF_COST);
+
+ format %{ "ClearArray $cnt, $base \t// KILL $tmp" %}
+ ins_encode %{
+ // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+ __ clear_memory_doubleword($base$$Register, $tmp$$Register, R0, $cnt$$constant); // kills base, R0
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+// Clear-array with dynamic array length.
instruct inlineCallClearArray(rarg1RegL cnt, rarg2RegP base, Universe dummy, regCTR ctr) %{
match(Set dummy (ClearArray cnt base));
effect(USE_KILL cnt, USE_KILL base, KILL ctr);
- ins_cost(MEMORY_REF_COST);
-
- ins_alignment(4); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
+ ins_cost(4 * MEMORY_REF_COST);
format %{ "ClearArray $cnt, $base" %}
ins_encode %{
// TODO: PPC port $archOpcode(ppc64Opcode_compound);
- __ clear_memory_doubleword($base$$Register, $cnt$$Register); // kills cnt, base, R0
+ __ clear_memory_doubleword($base$$Register, $cnt$$Register, R0); // kills cnt, base, R0
%}
ins_pipe(pipe_class_default);
%}
--- a/hotspot/src/share/vm/aot/aotLoader.cpp Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/src/share/vm/aot/aotLoader.cpp Mon Dec 19 00:49:34 2016 +0100
@@ -125,14 +125,18 @@
if (UseAOT) {
// EagerInitialization is not compatible with AOT
if (EagerInitialization) {
- warning("EagerInitialization is not compatible with AOT (switching AOT off)");
+ if (PrintAOT) {
+ warning("EagerInitialization is not compatible with AOT (switching AOT off)");
+ }
FLAG_SET_DEFAULT(UseAOT, false);
return;
}
// -Xint is not compatible with AOT
if (Arguments::is_interpreter_only()) {
- warning("-Xint is not compatible with AOT (switching AOT off)");
+ if (PrintAOT) {
+ warning("-Xint is not compatible with AOT (switching AOT off)");
+ }
FLAG_SET_DEFAULT(UseAOT, false);
return;
}
--- a/hotspot/test/Makefile Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/test/Makefile Mon Dec 19 00:49:34 2016 +0100
@@ -95,29 +95,6 @@
ALT_MAKE ?= closed
-include $(ALT_MAKE)/Makefile
-# Make sure jtreg exists
-$(JTREG): $(JT_HOME)
-
-jtreg_tests: prep $(PRODUCT_HOME) $(JTREG)
- ( \
- ( JT_HOME=$(shell $(GETMIXEDPATH) "$(JT_HOME)"); \
- export JT_HOME; \
- $(shell $(GETMIXEDPATH) "$(JTREG)") \
- $(JTREG_BASIC_OPTIONS) \
- -r:$(shell $(GETMIXEDPATH) "$(ABS_TEST_OUTPUT_DIR)/JTreport") \
- -w:$(shell $(GETMIXEDPATH) "$(ABS_TEST_OUTPUT_DIR)/JTwork") \
- -jdk:$(shell $(GETMIXEDPATH) "$(PRODUCT_HOME)") \
- $(JTREG_NATIVE_PATH) \
- $(JTREG_FAILURE_HANDLER_OPTIONS) \
- $(JTREG_EXCLUSIONS) \
- $(JTREG_TEST_OPTIONS) \
- $(TEST_SELECTION) \
- ) ; \
- $(BUNDLE_UP_AND_EXIT) \
- ) 2>&1 | $(TEE) $(ABS_TEST_OUTPUT_DIR)/output.txt ; $(TESTEXIT)
-
-PHONY_LIST += jtreg_tests
-
# flags used to execute java in test targets
TEST_FLAGS += -version -Xinternalversion -X -help
--- a/hotspot/test/compiler/aot/DeoptimizationTest.java Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/test/compiler/aot/DeoptimizationTest.java Mon Dec 19 00:49:34 2016 +0100
@@ -33,7 +33,9 @@
* @run main compiler.aot.AotCompiler -libname libDeoptimizationTest.so
* -class compiler.aot.DeoptimizationTest
* -compile compiler.aot.DeoptimizationTest.testMethod()D
+ * -extraopt -XX:-UseCompressedOops
* @run main/othervm -Xmixed -XX:+UseAOT -XX:+TieredCompilation
+ * -XX:-UseCompressedOops
* -XX:CompileCommand=dontinline,compiler.aot.DeoptimizationTest::*
* -XX:AOTLibrary=./libDeoptimizationTest.so -Xbootclasspath/a:.
* -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI
--- a/hotspot/test/compiler/aot/RecompilationTest.java Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/test/compiler/aot/RecompilationTest.java Mon Dec 19 00:49:34 2016 +0100
@@ -30,29 +30,36 @@
* compiler.aot.AotCompiler
* @run main ClassFileInstaller sun.hotspot.WhiteBox
* sun.hotspot.WhiteBox$WhiteBoxPermission
- * @run main compiler.aot.AotCompiler -libname libRecompilationTest.so
+ * @run main compiler.aot.AotCompiler -libname libRecompilationTest1.so
* -class compiler.whitebox.SimpleTestCaseHelper
* -extraopt -Dgraal.TieredAOT=true -extraopt -Dgraal.ProfileSimpleMethods=true
* -extraopt -XX:+UnlockDiagnosticVMOptions -extraopt -XX:+WhiteBoxAPI -extraopt -Xbootclasspath/a:.
- * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:+TieredCompilation -XX:-UseCounterDecay
- * -XX:CompileCommand=dontinline,*.*
- * -XX:AOTLibrary=./libRecompilationTest.so -Xbootclasspath/a:.
+ * -extraopt -XX:-UseCompressedOops
+ * -extraopt -XX:CompileCommand=dontinline,compiler.whitebox.SimpleTestCaseHelper::*
+ * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:+TieredCompilation
+ * -XX:-UseCounterDecay -XX:-UseCompressedOops
+ * -XX:-Inline
+ * -XX:AOTLibrary=./libRecompilationTest1.so -Xbootclasspath/a:.
* -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI
* -Dcompiler.aot.RecompilationTest.check_level=1
* compiler.aot.RecompilationTest
- * @run main compiler.aot.AotCompiler -libname libRecompilationTest.so
+ * @run main compiler.aot.AotCompiler -libname libRecompilationTest2.so
* -class compiler.whitebox.SimpleTestCaseHelper
* -extraopt -Dgraal.TieredAOT=false
* -extraopt -XX:+UnlockDiagnosticVMOptions -extraopt -XX:+WhiteBoxAPI -extraopt -Xbootclasspath/a:.
- * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:+TieredCompilation -XX:-UseCounterDecay
- * -XX:CompileCommand=dontinline,*.*
- * -XX:AOTLibrary=./libRecompilationTest.so -Xbootclasspath/a:.
+ * -extraopt -XX:-UseCompressedOops
+ * -extraopt -XX:CompileCommand=dontinline,compiler.whitebox.SimpleTestCaseHelper::*
+ * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:+TieredCompilation
+ * -XX:-UseCounterDecay -XX:-UseCompressedOops
+ * -XX:-Inline
+ * -XX:AOTLibrary=./libRecompilationTest2.so -Xbootclasspath/a:.
* -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI
* -Dcompiler.aot.RecompilationTest.check_level=-1
* compiler.aot.RecompilationTest
- * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:-TieredCompilation -XX:-UseCounterDecay
- * -XX:CompileCommand=dontinline,*.*
- * -XX:AOTLibrary=./libRecompilationTest.so -Xbootclasspath/a:.
+ * @run main/othervm -Xmixed -Xbatch -XX:+UseAOT -XX:-TieredCompilation
+ * -XX:-UseCounterDecay -XX:-UseCompressedOops
+ * -XX:-Inline
+ * -XX:AOTLibrary=./libRecompilationTest2.so -Xbootclasspath/a:.
* -XX:+UnlockDiagnosticVMOptions -XX:+WhiteBoxAPI
* -Dcompiler.aot.RecompilationTest.check_level=-1
* compiler.aot.RecompilationTest
--- a/hotspot/test/compiler/aot/SharedUsageTest.java Sun Dec 18 15:37:50 2016 +0100
+++ b/hotspot/test/compiler/aot/SharedUsageTest.java Mon Dec 19 00:49:34 2016 +0100
@@ -30,7 +30,9 @@
* compiler.aot.AotCompiler
* @run main compiler.aot.AotCompiler -libname libSharedUsageTest.so
* -class compiler.aot.SharedUsageTest
+ * -extraopt -XX:-UseCompressedOops
* @run main/othervm -XX:+UseAOT -XX:AOTLibrary=./libSharedUsageTest.so
+ * -XX:-UseCompressedOops
* -Dcompiler.aot.SharedUsageTest.parent=true
* compiler.aot.SharedUsageTest
* @summary check if .so can be successfully shared with 2 java processes