src/cpu/x86/vm/assembler_x86.cpp

Print this page
rev 3362 : 7174532: jdk/test/java/lang/Math/WorstCaseTests.java failing on x86
Summary: increase precision on x86 for the steps of the computation of exp and pow.
Reviewed-by:

*** 6925,6949 **** --- 6925,6970 ---- movl(Address(rsp,0),0); fmul_d(Address(rsp,0)); // Stack: 2^X ... addptr(rsp,sizeof(jdouble)); } + void MacroAssembler::increase_precision() { + subptr(rsp, BytesPerWord); + fnstcw(Address(rsp, 0)); + movl(rax, Address(rsp, 0)); + orl(rax, 0x300); + push(rax); + fldcw(Address(rsp, 0)); + pop(rax); + } + + void MacroAssembler::restore_precision() { + fldcw(Address(rsp, 0)); + addptr(rsp, BytesPerWord); + } + void MacroAssembler::fast_pow() { // computes X^Y = 2^(Y * log2(X)) // if fast computation is not possible, result is NaN. Requires // fallback from user of this macro. + // increase precision for intermediate steps of the computation + increase_precision(); fyl2x(); // Stack: (Y*log2(X)) ... pow_exp_core_encoding(); // Stack: exp(X) ... + restore_precision(); } void MacroAssembler::fast_exp() { // computes exp(X) = 2^(X * log2(e)) // if fast computation is not possible, result is NaN. Requires // fallback from user of this macro. + // increase precision for intermediate steps of the computation + increase_precision(); fldl2e(); // Stack: log2(e) X ... fmulp(1); // Stack: (X*log2(e)) ... pow_exp_core_encoding(); // Stack: exp(X) ... + restore_precision(); } void MacroAssembler::pow_or_exp(bool is_exp, int num_fpu_regs_in_use) { // kills rax, rcx, rdx // pow and exp needs 2 extra registers on the fpu stack.