Added opencl driver info to gpu_info.
Discarding DepthWise3x3 for specific driver.
PiperOrigin-RevId: 382682255
Change-Id: I07428e2d5bf55492ca614c0652e47c3f6cc20ed6
diff --git a/tensorflow/lite/delegates/gpu/cl/cl_device.cc b/tensorflow/lite/delegates/gpu/cl/cl_device.cc
index 10dd873..99b9b36 100644
--- a/tensorflow/lite/delegates/gpu/cl/cl_device.cc
+++ b/tensorflow/lite/delegates/gpu/cl/cl_device.cc
@@ -17,6 +17,7 @@
#include <algorithm>
#include <string>
+#include <utility>
#include <vector>
#include "absl/strings/numbers.h"
@@ -129,16 +130,21 @@
return gpu_version >= min_version && gpu_version < max_version;
}
-GpuInfo GpuInfoFromDeviceID(cl_device_id id) {
+GpuInfo GpuInfoFromDeviceID(cl_device_id id, cl_platform_id platform_id) {
GpuInfo info;
- const auto device_name = GetDeviceInfo<std::string>(id, CL_DEVICE_NAME);
- const auto vendor_name = GetDeviceInfo<std::string>(id, CL_DEVICE_VENDOR);
- const auto opencl_c_version =
+ info.opencl_info.platform_version =
+ GetPlatformInfo(platform_id, CL_PLATFORM_VERSION);
+ info.opencl_info.device_name = GetDeviceInfo<std::string>(id, CL_DEVICE_NAME);
+ info.opencl_info.vendor_name =
+ GetDeviceInfo<std::string>(id, CL_DEVICE_VENDOR);
+ info.opencl_info.opencl_c_version =
GetDeviceInfo<std::string>(id, CL_DEVICE_OPENCL_C_VERSION);
- const std::string gpu_description =
- absl::StrCat(device_name, " ", vendor_name, " ", opencl_c_version);
+ const std::string gpu_description = absl::StrCat(
+ info.opencl_info.device_name, " ", info.opencl_info.vendor_name, " ",
+ info.opencl_info.opencl_c_version);
GetGpuInfoFromDeviceDescription(gpu_description, GpuApi::kOpenCl, &info);
- info.opencl_info.cl_version = ParseCLVersion(opencl_c_version);
+ info.opencl_info.cl_version =
+ ParseCLVersion(info.opencl_info.opencl_c_version);
info.opencl_info.extensions =
absl::StrSplit(GetDeviceInfo<std::string>(id, CL_DEVICE_EXTENSIONS), ' ');
info.opencl_info.supports_fp16 = false;
@@ -261,7 +267,9 @@
} // namespace
CLDevice::CLDevice(cl_device_id id, cl_platform_id platform_id)
- : info_(GpuInfoFromDeviceID(id)), id_(id), platform_id_(platform_id) {
+ : info_(GpuInfoFromDeviceID(id, platform_id)),
+ id_(id),
+ platform_id_(platform_id) {
if (info_.IsAdreno() &&
info_.adreno_info.adreno_gpu == AdrenoGpu::kAdreno630) {
acceleration::AndroidInfo android_info;
diff --git a/tensorflow/lite/delegates/gpu/common/gpu_info.h b/tensorflow/lite/delegates/gpu/common/gpu_info.h
index e2adf1a..6d07d9c 100644
--- a/tensorflow/lite/delegates/gpu/common/gpu_info.h
+++ b/tensorflow/lite/delegates/gpu/common/gpu_info.h
@@ -264,6 +264,11 @@
std::string OpenClVersionToString(OpenClVersion version);
struct OpenClInfo {
+ std::string device_name;
+ std::string vendor_name;
+ std::string opencl_c_version;
+ std::string platform_version;
+
OpenClVersion cl_version;
std::vector<std::string> extensions;
diff --git a/tensorflow/lite/delegates/gpu/common/selectors/default/dw_convolution_selector.cc b/tensorflow/lite/delegates/gpu/common/selectors/default/dw_convolution_selector.cc
index e434e89..f4956e7 100644
--- a/tensorflow/lite/delegates/gpu/common/selectors/default/dw_convolution_selector.cc
+++ b/tensorflow/lite/delegates/gpu/common/selectors/default/dw_convolution_selector.cc
@@ -26,7 +26,7 @@
std::unique_ptr<GPUOperation> SelectDWConvolutionAdreno(
const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info,
const OperationDef& op_def) {
- if (IsDepthwiseConv3x3Supported(attr)) {
+ if (IsDepthwiseConv3x3Supported(gpu_info, attr)) {
return absl::make_unique<DepthwiseConv3x3>(
CreateDepthwiseConv3x3(gpu_info, op_def, attr));
} else {
@@ -38,7 +38,7 @@
std::unique_ptr<GPUOperation> SelectDWConvolutionPowerVR(
const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info,
const OperationDef& op_def) {
- if (IsDepthwiseConv3x3Supported(attr)) {
+ if (IsDepthwiseConv3x3Supported(gpu_info, attr)) {
return absl::make_unique<DepthwiseConv3x3>(
CreateDepthwiseConv3x3(gpu_info, op_def, attr));
} else {
@@ -54,7 +54,7 @@
bool buffer_type = storage_type == TensorStorageType::BUFFER ||
storage_type == TensorStorageType::IMAGE_BUFFER;
const MaliInfo mali_info = gpu_info.mali_info;
- if (IsDepthwiseConv3x3Supported(attr) && !mali_info.IsMidgard() &&
+ if (IsDepthwiseConv3x3Supported(gpu_info, attr) && !mali_info.IsMidgard() &&
!buffer_type && op_def.precision != CalculationsPrecision::F32) {
return absl::make_unique<DepthwiseConv3x3>(
CreateDepthwiseConv3x3(gpu_info, op_def, attr));
@@ -67,7 +67,7 @@
std::unique_ptr<GPUOperation> SelectDWConvolutionApple(
const DepthwiseConvolution2DAttributes& attr, const GpuInfo& gpu_info,
const OperationDef& op_def) {
- if (IsDepthwiseConv3x3Supported(attr)) {
+ if (IsDepthwiseConv3x3Supported(gpu_info, attr)) {
return absl::make_unique<DepthwiseConv3x3>(
CreateDepthwiseConv3x3(gpu_info, op_def, attr));
} else if (IsDepthWiseConv3x3StrideH2Supported(attr)) {
diff --git a/tensorflow/lite/delegates/gpu/common/tasks/BUILD b/tensorflow/lite/delegates/gpu/common/tasks/BUILD
index 95a69e0..391c561 100644
--- a/tensorflow/lite/delegates/gpu/common/tasks/BUILD
+++ b/tensorflow/lite/delegates/gpu/common/tasks/BUILD
@@ -444,6 +444,7 @@
"//tensorflow/lite/delegates/gpu/common/task:tensor_desc",
"//tensorflow/lite/delegates/gpu/common/task:texture2d_desc",
"//tensorflow/lite/delegates/gpu/common/task:work_group_picking",
+ "@com_google_absl//absl/strings",
],
)
diff --git a/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.cc b/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.cc
index 8358aca..2ff9605 100644
--- a/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.cc
+++ b/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.cc
@@ -18,6 +18,7 @@
#include <string>
#include <utility>
+#include "absl/strings/match.h"
#include "tensorflow/lite/delegates/gpu/common/status.h"
#include "tensorflow/lite/delegates/gpu/common/task/work_group_picking.h"
@@ -299,7 +300,16 @@
}
}
-bool IsDepthwiseConv3x3Supported(const DepthwiseConvolution2DAttributes& attr) {
+bool IsDepthwiseConv3x3Supported(const GpuInfo& gpu_info,
+ const DepthwiseConvolution2DAttributes& attr) {
+ if (gpu_info.IsApiOpenCl() && gpu_info.IsAdreno()) {
+ const std::string kBadDriver =
+ "OpenCL 2.0 QUALCOMM build: commit #7daed58 changeid #I7ece6fe30d "
+ "Date: 10/19/16";
+ if (absl::StrContains(gpu_info.opencl_info.platform_version, kBadDriver)) {
+ return false;
+ }
+ }
return attr.weights.shape.o == 1 && attr.dilations.w == 1 &&
attr.dilations.h == 1 && attr.weights.shape.w == 3 &&
attr.weights.shape.h == 3 && attr.strides.w == 1 &&
diff --git a/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.h b/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.h
index 59d1858..2793b1a 100644
--- a/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.h
+++ b/tensorflow/lite/delegates/gpu/common/tasks/depthwise_conv_3x3.h
@@ -146,7 +146,8 @@
}
}
-bool IsDepthwiseConv3x3Supported(const DepthwiseConvolution2DAttributes& attr);
+bool IsDepthwiseConv3x3Supported(const GpuInfo& gpu_info,
+ const DepthwiseConvolution2DAttributes& attr);
DepthwiseConv3x3 CreateDepthwiseConv3x3(
const GpuInfo& gpu_info, const OperationDef& definition,
