https://bugs.freedesktop.org/show_bug.cgi?id=91124
--- Comment #1 from Ilia Mirkin <[email protected]> --- Created attachment 116817 --> https://bugs.freedesktop.org/attachment.cgi?id=116817&action=edit shader that triggers bug The attached shader appears to trigger an issue in the FlatteningPass. Specifically the resulting code looks like 00000198: a0036003 00000000 joinat 0x1b0 000001a0: b00005fd 600187c8 set $c0 # ge f32 $r2 $r0 000001a8: 1000020d 0403c100 (e $c0) mov b32 $r3 $r1 000001b0: 10000801 2400c780 B ld $r0 b32 c0[0x10] That joinat shouldn't be there at all. This is a bit tricky, since pre-RA, the code looks like 60: joinat BB:5 (0) 61: set u8 %c259 ge f32 %r215 %r255 (0) 62: eq %c259 bra BB:4 (0) BB:3 (1 instructions) - idom = BB:12, df = { BB:5 } -> BB:5 (forward) 63: bra BB:5 (0) BB:4 (1 instructions) - idom = BB:12, df = { BB:5 } -> BB:5 (forward) 64: bra BB:5 (0) BB:5 (14 instructions) - idom = BB:12, df = { BB:10 } -> BB:7 (tree) -> BB:6 (tree) 65: phi u32 %r261 %r254 %r252 (0) 66: join (0) and the (e $c0) mov ... is actually generated from the phi node (since it creates constraint moves into BB:3/4. Post-RA but pre-FlatteningPass, this looks like 64: joinat BB:5 (0) 65: set u8 $c0 ge f32 $r2 $r0 (0) 66: eq $c0 bra BB:4 (0) BB:3 (1 instructions) - idom = BB:12, df = { BB:5 } -> BB:5 (forward) 67: bra BB:5 (0) BB:4 (2 instructions) - idom = BB:12, df = { BB:5 } -> BB:5 (forward) 68: mov u32 $r3 $r1 (0) 69: bra BB:5 (0) BB:5 (13 instructions) - idom = BB:12, df = { BB:10 } -> BB:7 (tree) -> BB:6 (tree) 70: join (0) Will have to understand the FlatteningPass better before I can figure out what's wrong though. -- You are receiving this mail because: You are the QA Contact for the bug. You are the assignee for the bug.
_______________________________________________ Nouveau mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/nouveau
