!8192 [MSLITE][Develop] fix bug of arm cpu fp16 op cast; delete arm cpu fp16 pad register

Merge pull request !8192 from yangruoqi713/lite

mindspore/lite/nnacl/fp16/cast_fp16.c

@@ -27,6 +27,18 @@ void Uint8ToFloat16(const uint8_t *input, float16_t *output, int number) {
   }
 }
 
+void Float16ToInt32(const float16_t *input, int32_t *output, int number) {
+  for (int i = 0; i < number; ++i) {
+    output[i] = (int32_t)input[i];
+  }
+}
+
+void Float16ToInt64(const float16_t *input, int64_t *output, int number) {
+  for (int i = 0; i < number; ++i) {
+    output[i] = (int64_t)input[i];
+  }
+}
+
 #ifndef ENABLE_ARM64
 void Float32ToFloat16(const float *input, float16_t *output, int number) {
   for (int i = 0; i < number; ++i) {

mindspore/lite/nnacl/fp16/cast_fp16.h

@@ -24,6 +24,8 @@ extern "C" {
 #endif
 void BoolToFloat16(const bool *input, float16_t *output, int number);
 void Uint8ToFloat16(const uint8_t *input, float16_t *output, int number);
+void Float16ToInt32(const float16_t *input, int32_t *output, int number);
+void Float16ToInt64(const float16_t *input, int64_t *output, int number);
 void Float32ToFloat16(const float *input, float16_t *output, int number);
 void Float16ToFloat32(const float16_t *input, float *output, int number);
 #ifdef __cplusplus

mindspore/lite/src/runtime/kernel/arm/fp16/cast_fp16.cc

@@ -65,25 +65,58 @@ int CastFp16CPUKernel::DoCast(int thread_id) {
   }
 
   auto offset = thread_id * stride_;
-  auto output_data = out_tensors_.at(0)->MutableData();
-  switch (input->data_type()) {
-    case kNumberTypeBool:
-      BoolToFloat16(reinterpret_cast<bool *>(input->MutableData()) + offset,
-                    reinterpret_cast<float16_t *>(output_data) + offset, data_num);
-    case kNumberTypeUInt8:
-      Uint8ToFloat16(reinterpret_cast<uint8_t *>(input->MutableData()) + offset,
-                     reinterpret_cast<float16_t *>(output_data) + offset, data_num);
-    case kNumberTypeFloat32:
-      Float32ToFloat16(reinterpret_cast<float *>(input->MutableData()) + offset,
-                       reinterpret_cast<float16_t *>(output_data) + offset, data_num);
-      break;
-    case kNumberTypeFloat16:
-      Float16ToFloat32(reinterpret_cast<float16_t *>(input->MutableData()) + offset,
-                       reinterpret_cast<float *>(output_data) + offset, data_num);
-      break;
-    default:
-      MS_LOG(ERROR) << "Unsupported input data type " << input->data_type();
-      return RET_ERROR;
+  auto output = out_tensors_.at(0);
+  auto output_data = output->data_c();
+  auto input_data_type = input->data_type();
+  auto output_data_type = output->data_type();
+
+  if (input_data_type == kNumberTypeFloat16) {
+    switch (output_data_type) {
+      case kNumberTypeInt64:
+        Float16ToInt64(reinterpret_cast<float16_t *>(input->data_c()) + offset,
+                       reinterpret_cast<int64_t *>(output_data) + offset, data_num);
+        break;
+      case kNumberTypeInt32:
+        Float16ToInt32(reinterpret_cast<float16_t *>(input->data_c()) + offset,
+                       reinterpret_cast<int32_t *>(output_data) + offset, data_num);
+        break;
+      case kNumberTypeFloat32:
+        Float16ToFloat32(reinterpret_cast<float16_t *>(input->MutableData()) + offset,
+                         reinterpret_cast<float *>(output_data) + offset, data_num);
+        break;
+      case kNumberTypeFloat16:
+        memcpy(reinterpret_cast<float16_t *>(output_data) + offset,
+               reinterpret_cast<float16_t *>(input->data_c()) + offset, data_num * sizeof(float16_t));
+        break;
+      default:
+        MS_LOG(ERROR) << "Unsupported output data type " << output_data_type;
+        return RET_ERROR;
+    }
+  } else if (input_data_type == kNumberTypeFloat32) {
+    switch (output_data_type) {
+      case kNumberTypeInt64:
+        Float32ToInt64(reinterpret_cast<float *>(input->data_c()) + offset,
+                       reinterpret_cast<int64_t *>(output_data) + offset, data_num);
+        break;
+      case kNumberTypeInt32:
+        Float32ToInt32(reinterpret_cast<float *>(input->data_c()) + offset,
+                       reinterpret_cast<int32_t *>(output_data) + offset, data_num);
+        break;
+      case kNumberTypeFloat32:
+        memcpy(reinterpret_cast<float *>(output_data) + offset, reinterpret_cast<float *>(input->data_c()) + offset,
+               data_num * sizeof(float));
+        break;
+      case kNumberTypeFloat16:
+        Float32ToFloat16(reinterpret_cast<float *>(input->MutableData()) + offset,
+                         reinterpret_cast<float16_t *>(output_data) + offset, data_num);
+        break;
+      default:
+        MS_LOG(ERROR) << "Unsupported output data type " << output_data_type;
+        return RET_ERROR;
+    }
+  } else {
+    MS_LOG(ERROR) << "Unsupported input data type " << input_data_type;
+    return RET_ERROR;
   }
   return RET_OK;
 }

mindspore/lite/src/runtime/kernel/arm/fp16/pad_fp16.cc

@@ -94,6 +94,4 @@ kernel::LiteKernel *CpuPadFp16KernelCreator(const std::vector<lite::Tensor *> &i
   }
   return kernel;
 }
-
-REG_KERNEL(kCPU, kNumberTypeFloat16, PrimitiveType_Pad, CpuPadFp16KernelCreator)
 }  // namespace mindspore::kernel

mindspore/lite/src/scheduler.cc

@@ -304,7 +304,8 @@ TypeId Scheduler::GetFirstFp32Fp16OrInt8Type(const std::vector<Tensor *> &in_ten
       return dtype;
     }
   }
-  return kNumberTypeFloat32;
+  MS_ASSERT(in_tensors.size() > 0);
+  return in_tensors[0]->data_type();
 }
 
 void Scheduler::SetKernelTensorDataType(kernel::LiteKernel *kernel) {
@@ -346,7 +347,8 @@ kernel::SubGraphType Scheduler::GetKernelSubGraphType(kernel::LiteKernel *kernel
     if (desc.data_type == kNumberTypeFloat16) {
       return kernel::kCpuFP16SubGraph;
     } else if (desc.data_type == kNumberTypeFloat32 || desc.data_type == kNumberTypeInt8 ||
-               desc.data_type == kNumberTypeInt32 || desc.data_type == kNumberTypeBool) {
+               desc.data_type == kNumberTypeInt32 || desc.data_type == kNumberTypeBool ||
+               desc.data_type == kNumberTypeUInt8) {
       return kernel::kCpuFP32SubGraph;
     }
   }