for linux 2.4.3 in arch/alpha/math-emu/math ( approx line 215 ) have a try with this
change. This should fix a conversion of a single FP denorm, into a double.
/gat
\
> case FOP_FNC_CVTxS:
> /* It is irritating that DEC encoded CVTST with
> SRC == T_floating. It is also interesting that
> the bit used to tell the two apart is /U... */
> if (insn & 0x2000) {
> FP_CONV(S,D,1,1,SR,DB);
> goto pack_s;
> } else {
> /* CVTST need do nothing else but copy the
> bits and repack. */
> /* Not exactly,
> * Conversion is from single -> double
> * so vb needs to be in a 32 bit
> * float form, and UNPACK'ed as such! */
> vb = alpha_read_fp_reg_s(fb);
> FP_UNPACK_SP(SB, &vb);
> DR_c = SB_c;
> DR_s = SB_s;
> DR_e = SB_e;
> DR_f = SB_f << ( 52 - 23 );
>
> goto pack_d;
> }
>
>
>
>
> The problem I'm running into is when denorms are generated within optimized
> libraries like CXML (specifically the BLAS routines). I've tried to
> c
_______________________________________________
Axp-list mailing list
Axp-list@redhat.com
https://listman.redhat.com/mailman/listinfo/axp-list
This archive was generated by hypermail version 2a22 on Sat May 5 06:18:14 2001 PDT
Send any problems or questions about this archive to webmaster@alphalinux.org.