2 * use stfiwx in float->int
3 * implement cast fp to bool
4 * implement algebraic shift right long by reg
5 * implement scheduling info
6 * implement powerpc-64 for darwin
7 * implement powerpc-64 for aix
8 * fix rlwimi generation to be use-and-def
10 floatdidf assumes signed longs. so if the high but of a ulong
11 just happens to be set, you get the wrong sign. The fix for this
12 is to call cmpdi2 to compare against zero, if so shift right by one,
13 convert to fp, and multiply by (add to itself). the sequence would
15 {r3:r4} holds ulong a;
17 li r6, 0 (set r5:r6 to ulong 0)
18 call cmpdi2 ==> sets r3 <, =, > 0
20 call floatdidf as usual
22 shift right ulong a, 1 (we could use emitShift)
24 fadd f1, f1, f1 (fp left shift by 1)
25 * cast elimination pass (uint -> sbyte -> short, kill the byte -> short)
26 * should hint to the branch select pass that it doesn't need to print the
27 second unconditional branch, so we don't end up with things like:
28 b .LBBl42__2E_expand_function_8_674 ; loopentry.24
29 b .LBBl42__2E_expand_function_8_42 ; NewDefault
30 b .LBBl42__2E_expand_function_8_42 ; NewDefault
32 Currently failing tests that should pass:
35 | `- casts (ulong to fp failure)
38 | `- hbd: miscompilation