ThoughtWorksInc · Atry · May 9, 2018 · May 9, 2018
diff --git a/OpenCL/src/main/scala/com/thoughtworks/compute/OpenCL.scala b/OpenCL/src/main/scala/com/thoughtworks/compute/OpenCL.scala
@@ -929,7 +929,7 @@ object OpenCL {
       checkBuildErrorCode(None, clBuildProgram(handle, null, options, null, NULL))
     }
 
-    def build(): Unit = build("")
+    def build()(implicit witness: Witness.Aux[Owner]): Unit = build(witness.value.defaultProgramOptions)
 
     def monadicClose = UnitContinuation.delay {
       OpenCL.checkErrorCode(clReleaseProgram(handle))
@@ -1113,11 +1113,34 @@ object OpenCL {
     }
   }
 
+  /** A plug-in of Tensors to suppress warnings during compiling a OpenCL kernel for non-AMD platforms. */
+  trait SuppressWarnings extends OpenCL {
+    @transient
+    private lazy val _defaultProgramOptions = {
+      if (platformCapabilities.cl_amd_compile_options) {
+        // AMD SDK does not support -w flag in OpenCL specification.
+        super.defaultProgramOptions
+      } else {
+        super.defaultProgramOptions + " -w"
+      }
+    }
+
+    override protected def defaultProgramOptions: CharSequence = _defaultProgramOptions
+  }
+
+  trait UnsafeMathOptimizations extends OpenCL {
+    private lazy val _defaultProgramOptions = super.defaultProgramOptions + " -cl-unsafe-math-optimizations"
+
+    abstract override protected def defaultProgramOptions: CharSequence = _defaultProgramOptions
+  }
+
 }
 
 trait OpenCL extends MonadicCloseable[UnitContinuation] with DefaultCloseable {
   import OpenCL._
 
+  protected def defaultProgramOptions: CharSequence = ""
+
   protected def createKernels(program: Program): Seq[Kernel] = {
     val stack = stackPush()
     try {

diff --git a/Tensors/src/main/scala/com/thoughtworks/compute/Tensors.scala b/Tensors/src/main/scala/com/thoughtworks/compute/Tensors.scala
@@ -41,26 +41,6 @@ object Tensors {
 
   private final val MaxWorkItemDimensions = 3
 
-  /** A plug-in of Tensors to suppress warnings during compiling a OpenCL kernel for non-AMD platforms. */
-  trait SuppressWarnings extends Tensors {
-    @transient
-    private lazy val _openclCompilerFlags = {
-      if (platformCapabilities.cl_amd_compile_options) {
-        // AMD SDK does not support -w flag in OpenCL specification.
-        super.openclCompilerFlags
-      } else {
-        super.openclCompilerFlags + " -w"
-      }
-    }
-
-    override protected def openclCompilerFlags: String = _openclCompilerFlags
-  }
-
-  trait UnsafeMathOptimizations extends Tensors {
-    private lazy val _openclCompilerFlags = super.openclCompilerFlags + " -cl-unsafe-math-optimizations"
-    override protected def openclCompilerFlags: String = _openclCompilerFlags
-  }
-
   trait TensorBuilder[Data] {
     type Element
     def flatten(a: Data): Seq[Element]
@@ -320,8 +300,6 @@ trait Tensors extends OpenCL {
 
   protected def hashSourceCode: Fastring
 
-  protected def openclCompilerFlags: String = ""
-
   protected object PlusPrograms extends MonoidPrograms {
     def append(leftHandSide: Fastring, rightHandSide: Fastring): Fastring = fast"(($leftHandSide) + ($rightHandSide))"
     def zero: Fastring = fast"0.0f"
@@ -368,7 +346,7 @@ trait Tensors extends OpenCL {
           }
         }
       """)
-      program.build(openclCompilerFlags)
+      program.build()
       program
     }
 
@@ -409,7 +387,7 @@ trait Tensors extends OpenCL {
           }
         }
       """)
-      program.build(openclCompilerFlags)
+      program.build()
       program
     }
   }
@@ -446,7 +424,7 @@ trait Tensors extends OpenCL {
           buffer[i * 2 + 1] = z1;
         }
       """)
-      program.build(openclCompilerFlags)
+      program.build()
       program
     }
 
@@ -460,7 +438,7 @@ trait Tensors extends OpenCL {
       buffer[i] = hash(i ^ seed) / 4294967296.0f;
     }
     """)
-      program.build(openclCompilerFlags)
+      program.build()
       program
     }
 
@@ -1049,7 +1027,7 @@ trait Tensors extends OpenCL {
     /**
       * @group delayed
       */
-    def transpose: TransformedTensor = { permute(shape.indices.reverse.toArray)}
+    def transpose: TransformedTensor = { permute(shape.indices.reverse.toArray) }
 
     /**
       * @group delayed
@@ -1216,7 +1194,7 @@ trait Tensors extends OpenCL {
               }
 
               val program = createProgramWithSource(sourceCode)
-              program.build(openclCompilerFlags)
+              program.build()
 
               val compiledKernel = new CompiledKernel {
 

diff --git a/benchmarks/src/jmh/scala/com/thoughtworks/compute/benchmarks.scala b/benchmarks/src/jmh/scala/com/thoughtworks/compute/benchmarks.scala
@@ -33,8 +33,8 @@ object benchmarks {
 
     trait BenchmarkTensors
         extends StrictLogging
-        with Tensors.UnsafeMathOptimizations
-        with Tensors.SuppressWarnings
+        with OpenCL.UnsafeMathOptimizations
+        with OpenCL.SuppressWarnings
         with OpenCL.LogContextNotification
         with OpenCL.UseAllDevicesByType
         with OpenCL.GlobalExecutionContext

diff --git a/cpu/src/main/scala/com/thoughtworks/compute/cpu.scala b/cpu/src/main/scala/com/thoughtworks/compute/cpu.scala
@@ -102,7 +102,7 @@ import org.lwjgl.opencl.CL10.CL_DEVICE_TYPE_CPU
   */
 object cpu
     extends StrictLogging
-    with Tensors.UnsafeMathOptimizations
+    with OpenCL.UnsafeMathOptimizations
     with OpenCL.LogContextNotification
     with OpenCL.GlobalExecutionContext
     with OpenCL.CommandQueuePool

diff --git a/gpu/src/main/scala/com/thoughtworks/compute/gpu.scala b/gpu/src/main/scala/com/thoughtworks/compute/gpu.scala
@@ -14,7 +14,7 @@ import com.typesafe.scalalogging.StrictLogging
   */
 object gpu
     extends StrictLogging
-    with Tensors.UnsafeMathOptimizations
+    with OpenCL.UnsafeMathOptimizations
     with OpenCL.LogContextNotification
     with OpenCL.GlobalExecutionContext
     with OpenCL.CommandQueuePool