DrSocalkwe3n/luajit
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Disable FMA by default. Use -Ofma or jit.opt.start("+fma") to enable.

Browse Source 
See the discussion in #918 for the rationale.
This commit is contained in:
Mike Pall
2022-12-07 18:38:22 +01:00
parent 7d5d4a1b1a
commit de2e1ca9d3
7 changed files with 35 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/lj_vmmath.c
View File

@@ -36,6 +36,17 @@ LJ_FUNCA double lj_wrap_fmod(double x, double y) { return fmod(x, y); }
/* -- Helper functions ---------------------------------------------------- */
/* Required to prevent the C compiler from applying FMA optimizations.
**
** Yes, there's -ffp-contract and the FP_CONTRACT pragma ... in theory.
** But the current state of C compilers is a mess in this regard.
** Also, this function is not performance sensitive at all.
*/
LJ_NOINLINE static double lj_vm_floormul(double x, double y)
{
return lj_vm_floor(x / y) * y;
}
double lj_vm_foldarith(double x, double y, int op)
{
switch (op) {
@@ -43,7 +54,7 @@ double lj_vm_foldarith(double x, double y, int op)
case IR_SUB - IR_ADD: return x-y; break;
case IR_MUL - IR_ADD: return x*y; break;
case IR_DIV - IR_ADD: return x/y; break;
case IR_MOD - IR_ADD: return x-lj_vm_floor(x/y)*y; break;
case IR_MOD - IR_ADD: return x-lj_vm_floormul(x, y); break;
case IR_POW - IR_ADD: return pow(x, y); break;
case IR_NEG - IR_ADD: return -x; break;
case IR_ABS - IR_ADD: return fabs(x); break;