add _SetDataRand float16 on GPU

source/tensor/core/getandset/SetData.cu

@@ -476,40 +476,28 @@ void _CudaSetDataRand(XTensor * tensor, DTYPE lower, DTYPE upper)
     curandGenerator_t gen;
     curandGenerator_t gen;
     curandCreateGenerator (&gen, CURAND_RNG_PSEUDO_DEFAULT);
     curandCreateGenerator (&gen, CURAND_RNG_PSEUDO_DEFAULT);
     curandSetPseudoRandomGeneratorSeed(gen, time(NULL));
     curandSetPseudoRandomGeneratorSeed(gen, time(NULL));
-    curandGenerateUniform(gen , (float*)tensor->data , tensor->unitNum);
-    curandDestroyGenerator(gen);
     DTYPE variance = upper - lower;
     DTYPE variance = upper - lower;
 
 
-    if (tensor->dataType == X_FLOAT) {
-        curandGenerator_t gen;
-        curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT);
-        curandSetPseudoRandomGeneratorSeed(gen, time(NULL));
-        curandGenerateUniform(gen, (float*)tensor->data, tensor->unitNum);
-        curandDestroyGenerator(gen);
-        DTYPE variance = upper - lower;
-        KernelSetDataRandFloat << <blocks, threads >> > ((float*)tensor->data, tensor->unitNum, lower, variance);
-    }
-    else if (tensor->dataType == X_DOUBLE) {
-        curandGenerator_t gen;
-        curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT);
-        curandSetPseudoRandomGeneratorSeed(gen, time(NULL));
-        curandGenerateUniform(gen, (float*)tensor->data, tensor->unitNum);
-        curandDestroyGenerator(gen);
-        DTYPE variance = upper - lower;
-        KernelSetDataRandDouble << <blocks, threads >> > ((double*)tensor->data, tensor->unitNum, lower, variance);
-    }
-    else if (tensor->dataType == X_FLOAT16) {
-        XTensor tensor1(tensor->order, tensor->dimSize, X_FLOAT, tensor->denseRatio, tensor->devID, tensor->mem);
-        curandGenerator_t gen;
-        curandCreateGenerator(&gen, CURAND_RNG_PSEUDO_DEFAULT);
-        curandSetPseudoRandomGeneratorSeed(gen, time(NULL));
-        curandGenerateUniform(gen, (float*)tensor1.data, tensor1.unitNum);
-        curandDestroyGenerator(gen);
-        _ConvertDataType(&tensor1, tensor);
-        DTYPE variance = upper - lower;
-        KernelSetDataRandHalf << <blocks, threads >> >((half*)tensor->data, tensor->unitNum, lower, variance);
-    }
+	if (tensor->dataType == X_FLOAT) {
+		curandGenerateUniform(gen, (float*)tensor->data, tensor->unitNum);
+		curandDestroyGenerator(gen);
+		KernelSetDataRandFloat << <blocks, threads >> >((float*)tensor->data, tensor->unitNum, lower, variance);
+	}
         
         
+	else if (tensor->dataType == X_DOUBLE) {
+		curandGenerateUniform(gen, (float*)tensor->data, tensor->unitNum);
+		curandDestroyGenerator(gen);
+		KernelSetDataRandDouble << <blocks, threads >> >((double*)tensor->data, tensor->unitNum, lower, variance);
+	}
+
+	else if (tensor->dataType == X_FLOAT16) {
+		XTensor tensor1(tensor->order, tensor->dimSize, X_FLOAT, tensor->denseRatio, tensor->devID,tensor->mem);
+		curandGenerateUniform(gen, (float *)tensor1.data, tensor1.unitNum);
+		curandDestroyGenerator(gen);
+		_ConvertDataType(&tensor1, tensor);
+		KernelSetDataRandHalf << <blocks, threads >> >((half*)tensor->data, tensor->unitNum, lower, variance);
+	}
+		
     BacktoCudaDev(tensor->devID, devIDBackup);
     BacktoCudaDev(tensor->devID, devIDBackup);
 }
 }