stan/jdk-24
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

8256061: RegisterSaver::save_live_registers() omits upper halves of ZMM0-15 registers

Browse Source 
Reviewed-by: kvn
This commit is contained in:
Vladimir Ivanov 2020-11-20 21:15:58 +00:00
parent 503590f6bf
commit f79e9d4573
1 changed files with 23 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp
View File

@ -89,9 +89,10 @@ class RegisterSaver {
#define XSAVE_AREA_YMM_BEGIN 576
#define XSAVE_AREA_ZMM_BEGIN 1152
#define XSAVE_AREA_UPPERBANK 1664
#define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
#define DEF_YMM_OFFS(regnum) ymm ## regnum ## _off = ymm_off + (regnum)*16/BytesPerInt, ymm ## regnum ## H_off
#define DEF_ZMM_OFFS(regnum) zmm ## regnum ## _off = zmm_off + (regnum-16)*64/BytesPerInt, zmm ## regnum ## H_off
#define DEF_XMM_OFFS(regnum) xmm ## regnum ## _off = xmm_off + (regnum)*16/BytesPerInt, xmm ## regnum ## H_off
#define DEF_YMM_OFFS(regnum) ymm ## regnum ## _off = ymm_off + (regnum)*16/BytesPerInt, ymm ## regnum ## H_off
#define DEF_ZMM_OFFS(regnum) zmm ## regnum ## _off = zmm_off + (regnum)*32/BytesPerInt, zmm ## regnum ## H_off
#define DEF_ZMM_UPPER_OFFS(regnum) zmm ## regnum ## _off = zmm_upper_off + (regnum-16)*64/BytesPerInt, zmm ## regnum ## H_off
enum layout {
fpu_state_off = frame::arg_reg_save_area_bytes/BytesPerInt, // fxsave save area
xmm_off = fpu_state_off + XSAVE_AREA_BEGIN/BytesPerInt, // offset in fxsave save area
@ -102,10 +103,12 @@ class RegisterSaver {
DEF_YMM_OFFS(0),
DEF_YMM_OFFS(1),
// 2..15 are implied in range usage
zmm_high = xmm_off + (XSAVE_AREA_ZMM_BEGIN - XSAVE_AREA_BEGIN)/BytesPerInt,
zmm_off = xmm_off + (XSAVE_AREA_UPPERBANK - XSAVE_AREA_BEGIN)/BytesPerInt,
DEF_ZMM_OFFS(16),
DEF_ZMM_OFFS(17),
zmm_off = xmm_off + (XSAVE_AREA_ZMM_BEGIN - XSAVE_AREA_BEGIN)/BytesPerInt,
DEF_ZMM_OFFS(0),
DEF_ZMM_OFFS(1),
zmm_upper_off = xmm_off + (XSAVE_AREA_UPPERBANK - XSAVE_AREA_BEGIN)/BytesPerInt,
DEF_ZMM_UPPER_OFFS(16),
DEF_ZMM_UPPER_OFFS(17),
// 18..31 are implied in range usage
fpu_state_end = fpu_state_off + ((FPUStateSizeInWords-1)*wordSize / BytesPerInt),
fpu_stateH_end,
@ -259,7 +262,7 @@ OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_
map->set_callee_saved(STACK_OFFSET(off), xmm_name->as_VMReg());
off += delta;
}
if(UseAVX > 2) {
if (UseAVX > 2) {
// Obtain xmm16..xmm31 from the XSAVE area on EVEX enabled targets
off = zmm16_off;
delta = zmm17_off - off;
@ -272,13 +275,24 @@ OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_
#if COMPILER2_OR_JVMCI
if (save_vectors) {
// Save upper half of YMM registers(0..15)
off = ymm0_off;
int delta = ymm1_off - off;
delta = ymm1_off - ymm0_off;
for (int n = 0; n < 16; n++) {
XMMRegister ymm_name = as_XMMRegister(n);
map->set_callee_saved(STACK_OFFSET(off), ymm_name->as_VMReg()->next(4));
off += delta;
}
if (VM_Version::supports_evex()) {
// Save upper half of ZMM registers(0..15)
off = zmm0_off;
delta = zmm1_off - zmm0_off;
for (int n = 0; n < 16; n++) {
XMMRegister zmm_name = as_XMMRegister(n);
map->set_callee_saved(STACK_OFFSET(off), zmm_name->as_VMReg()->next(8));
off += delta;
}
}
}
#endif // COMPILER2_OR_JVMCI