infernap12/jdk
1
0
Fork 0
mirror of https://github.com/openjdk/jdk.git synced 2026-04-13 08:30:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

8078113: 8011102 changes may cause incorrect results

Browse Source 
Replace Vzeroupper instruction in stubs with zeroing only used ymm registers.

Reviewed-by: kvn
This commit is contained in:
Sandhya Viswanathan 2015-04-17 17:39:19 -07:00 committed by Vladimir Kozlov
parent 22425d93ca
commit 51fd716978
3 changed files with 17 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
hotspot/src/cpu/x86/vm/macroAssembler_x86.cpp
View File

@ -6657,7 +6657,7 @@ void MacroAssembler::string_compare(Register str1, Register str2,
subl(cnt2, stride2);
jccb(Assembler::notZero, COMPARE_WIDE_VECTORS_LOOP);
// clean upper bits of YMM registers
vzeroupper();
vpxor(vec1, vec1);
// compare wide vectors tail
bind(COMPARE_WIDE_TAIL);
@ -6672,7 +6672,7 @@ void MacroAssembler::string_compare(Register str1, Register str2,
// Identifies the mismatching (higher or lower)16-bytes in the 32-byte vectors.
bind(VECTOR_NOT_EQUAL);
// clean upper bits of YMM registers
vzeroupper();
vpxor(vec1, vec1);
lea(str1, Address(str1, result, scale));
lea(str2, Address(str2, result, scale));
jmp(COMPARE_16_CHARS);
@ -6931,7 +6931,8 @@ void MacroAssembler::char_arrays_equals(bool is_array_equ, Register ary1, Regist
bind(DONE);
if (UseAVX >= 2) {
// clean upper bits of YMM registers
vzeroupper();
vpxor(vec1, vec1);
vpxor(vec2, vec2);
}
}
@ -7065,7 +7066,8 @@ void MacroAssembler::generate_fill(BasicType t, bool aligned,
BIND(L_check_fill_8_bytes);
// clean upper bits of YMM registers
vzeroupper();
movdl(xtmp, value);
pshufd(xtmp, xtmp, 0);
} else {
// Fill 32-byte chunks
pshufd(xtmp, xtmp, 0);
@ -7228,7 +7230,11 @@ void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
bind(L_copy_16_chars_exit);
if (UseAVX >= 2) {
// clean upper bits of YMM registers
vzeroupper();
vpxor(tmp2Reg, tmp2Reg);
vpxor(tmp3Reg, tmp3Reg);
vpxor(tmp4Reg, tmp4Reg);
movdl(tmp1Reg, tmp5);
pshufd(tmp1Reg, tmp1Reg, 0);
}
subptr(len, 8);
jccb(Assembler::greater, L_copy_8_chars_exit);

3
hotspot/src/cpu/x86/vm/stubGenerator_x86_32.cpp
View File

@ -835,7 +835,8 @@ class StubGenerator: public StubCodeGenerator {
if (UseUnalignedLoadStores && (UseAVX >= 2)) {
// clean upper bits of YMM registers
__ vzeroupper();
__ vpxor(xmm0, xmm0);
__ vpxor(xmm1, xmm1);
}
__ addl(qword_count, 8);
__ jccb(Assembler::zero, L_exit);

6
hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp
View File

@ -1352,7 +1352,8 @@ class StubGenerator: public StubCodeGenerator {
__ BIND(L_end);
if (UseAVX >= 2) {
// clean upper bits of YMM registers
__ vzeroupper();
__ vpxor(xmm0, xmm0);
__ vpxor(xmm1, xmm1);
}
} else {
// Copy 32-bytes per iteration
@ -1429,7 +1430,8 @@ class StubGenerator: public StubCodeGenerator {
__ BIND(L_end);
if (UseAVX >= 2) {
// clean upper bits of YMM registers
__ vzeroupper();
__ vpxor(xmm0, xmm0);
__ vpxor(xmm1, xmm1);
}
} else {
// Copy 32-bytes per iteration