From a27548400ea02c39b6602526eb697c673c7d22bb Mon Sep 17 00:00:00 2001
From: Ilia Mirkin <imirkin@alum.mit.edu>
Date: Tue, 8 Dec 2015 23:55:18 -0500
Subject: [PATCH] nv50/ir: fix assumption that prog->maxGPR is in 32-bit reg
 units

On NV50, we use 16-bit reg units (to make it all work with half-regs). A
few places assumed that it was always in 32-bit units.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
---
 .../nouveau/codegen/nv50_ir_lowering_nv50.cpp |  2 +-
 .../nouveau/codegen/nv50_ir_peephole.cpp      | 23 ++++++++++++++++---
 2 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
index 64f5fc0031e..82f7a8f330e 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
@@ -1225,7 +1225,7 @@ NV50LoweringPreSSA::handleEXPORT(Instruction *i)
          i->setDef(0, new_LValue(func, FILE_GPR));
          i->getDef(0)->reg.data.id = id;
 
-         prog->maxGPR = MAX2(prog->maxGPR, id);
+         prog->maxGPR = MAX2(prog->maxGPR, id * 2);
       }
    }
    return true;
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index 805be5fcb40..834fc7d7e30 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -2532,6 +2532,7 @@ MemoryOpt::runOpt(BasicBlock *bb)
 class FlatteningPass : public Pass
 {
 private:
+   virtual bool visit(Function *);
    virtual bool visit(BasicBlock *);
 
    bool tryPredicateConditional(BasicBlock *);
@@ -2540,6 +2541,8 @@ private:
    inline bool isConstantCondition(Value *pred);
    inline bool mayPredicate(const Instruction *, const Value *pred) const;
    inline void removeFlow(Instruction *);
+
+   uint8_t gpr_unit;
 };
 
 bool
@@ -2561,9 +2564,15 @@ FlatteningPass::isConstantCondition(Value *pred)
          file = ld->src(0).getFile();
       } else {
          file = insn->src(s).getFile();
-         // catch $r63 on NVC0
-         if (file == FILE_GPR && insn->getSrc(s)->reg.data.id > prog->maxGPR)
-            file = FILE_IMMEDIATE;
+         // catch $r63 on NVC0 and $r63/$r127 on NV50. Unfortunately maxGPR is
+         // in register "units", which can vary between targets.
+         if (file == FILE_GPR) {
+            Value *v = insn->getSrc(s);
+            int bytes = v->reg.data.id * MIN2(v->reg.size, 4);
+            int units = bytes >> gpr_unit;
+            if (units > prog->maxGPR)
+               file = FILE_IMMEDIATE;
+         }
       }
       if (file != FILE_IMMEDIATE && file != FILE_MEMORY_CONST)
          return false;
@@ -2668,6 +2677,14 @@ FlatteningPass::tryPropagateBranch(BasicBlock *bb)
    }
 }
 
+bool
+FlatteningPass::visit(Function *fn)
+{
+   gpr_unit = prog->getTarget()->getFileUnit(FILE_GPR);
+
+   return true;
+}
+
 bool
 FlatteningPass::visit(BasicBlock *bb)
 {
-- 
2.30.2