[Mesa-dev] [PATCH v3 3/9] nv50/ir: teach load propagation about src2

[Samuel Pitoiset]

With OP_ADD3, we might want to swap sources 2 and 1.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
---
 .../drivers/nouveau/codegen/nv50_ir_peephole.cpp   | 29 ++++++++++++++++++++++
 1 file changed, 29 insertions(+)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index a9172f8..f212eba 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -153,6 +153,7 @@ private:
    virtual bool visit(BasicBlock *);
 
    void checkSwapSrc01(Instruction *);
+   void checkSwapSrc21(Instruction *);
 
    bool isCSpaceLoad(Instruction *);
    bool isImmdLoad(Instruction *);
@@ -239,6 +240,32 @@ LoadPropagation::checkSwapSrc01(Instruction *insn)
    }
 }
 
+void
+LoadPropagation::checkSwapSrc21(Instruction *insn)
+{
+   const Target *targ = prog->getTarget();
+   if (insn->op != OP_ADD3)
+      return;
+   if (insn->src(2).getFile() != FILE_GPR)
+      return;
+
+   Instruction *i1 = insn->getSrc(1)->getInsn();
+   Instruction *i2 = insn->getSrc(2)->getInsn();
+
+   // Swap sources to inline the less frequently used source. That way,
+   // optimistically, it will eventually be able to remove the instruction.
+   int i1refs = insn->getSrc(1)->refCount();
+   int i2refs = insn->getSrc(2)->refCount();
+
+   if ((isCSpaceLoad(i2) || isImmdLoad(i2)) && targ->insnCanLoad(insn, 1, i2)) {
+      if ((!isImmdLoad(i1) && !isCSpaceLoad(i1)) ||
+          !targ->insnCanLoad(insn, 1, i1) ||
+          i2refs < i1refs) {
+         insn->swapSources(2, 1);
+      }
+   }
+}
+
 bool
 LoadPropagation::visit(BasicBlock *bb)
 {
@@ -256,6 +283,8 @@ LoadPropagation::visit(BasicBlock *bb)
 
       if (i->srcExists(1))
          checkSwapSrc01(i);
+      if (i->srcExists(2))
+         checkSwapSrc21(i);
 
       for (int s = 0; i->srcExists(s); ++s) {
          Instruction *ld = i->getSrc(s)->getInsn();
-- 
2.9.3