[Mesa-dev] [PATCH 1/2] nv50/ir: deal with loops with no breaks

[Ilia Mirkin]

For example if there are only returns, the break bb will not end up part
of the CFG. However there will have been a prebreak already emitted for
it, and when hitting the RET that comes after, we will try to insert the
current (i.e. break) BB into the graph even though it will be
unreachable. This makes the SSA code sad.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
Cc: "11.0 11.1" <mesa-stable at lists.freedesktop.org>
---
 src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
index 08a73d7..1d2caab 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -2893,6 +2893,12 @@ Converter::handleInstruction(const struct tgsi_full_instruction *insn)
          bb->cfg.attach(&loopBB->cfg, Graph::Edge::BACK);
       }
       setPosition(reinterpret_cast<BasicBlock *>(breakBBs.pop().u.p), true);
+
+      // If the loop never breaks (e.g. only has RET's inside), then there
+      // will be no way to get to the break bb. However BGNLOOP will have
+      // already made a PREBREAK to it, so it must be in the CFG.
+      if (getBB()->cfg.incidentCount() == 0)
+         loopBB->cfg.attach(&getBB()->cfg, Graph::Edge::TREE);
    }
       break;
    case TGSI_OPCODE_BRK:
-- 
2.4.10