nv50/ir/opt: improve post-multiply and check target for support

2 files changed, 13 insertions, 0 deletions

diff --git a/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.cpp b/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.cpp
index bcc2c43a55f..6fe95c6693f 100644
--- a/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.cpp
+++ b/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.cpp
@@ -516,6 +516,18 @@ TargetNVC0::isSatSupported(const Instruction *insn) const
    return insn->dType == TYPE_F32;
 }
 
+bool
+TargetNVC0::isPostMultiplySupported(operation op, float f, int& e) const
+{
+   if (op != OP_MUL)
+      return false;
+   f = fabsf(f);
+   e = static_cast<int>(log2f(f));
+   if (e < -3 || e > 3)
+      return false;
+   return f == exp2f(static_cast<float>(e));
+}
+
 // TODO: better values
 int TargetNVC0::getLatency(const Instruction *i) const
 {
diff --git a/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.h b/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.h
index 1a49f369920..e4efe476de0 100644
--- a/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.h
+++ b/src/gallium/drivers/nvc0/codegen/nv50_ir_target_nvc0.h
@@ -48,6 +48,7 @@ public:
    virtual bool isAccessSupported(DataFile, DataType) const;
    virtual bool isModSupported(const Instruction *, int s, Modifier) const;
    virtual bool isSatSupported(const Instruction *) const;
+   virtual bool isPostMultiplySupported(operation, float, int& e) const;
    virtual bool mayPredicate(const Instruction *, const Value *) const;
 
    virtual int getLatency(const Instruction *) const;