issue/291/refactor: 改造 +-*

Signed-off-by: YdrMaster <ydrml@hotmail.com>

src/infiniop/ops/add/cuda/add_cuda.cu → src/infiniop/ops/add/nvidia/add_nvidia.cu

-#include "add_cuda.cuh"
-#include "add_cuda_internal.cuh"
+#include "add_nvidia.cuh"
+#include "../cuda/kernel.cuh"
 
-namespace op::add::cuda {
+namespace op::add::nvidia {
 
 Descriptor::~Descriptor() = default;
 
@@ -43,13 +43,13 @@ infiniStatus_t Descriptor::calculate(
 
     switch (_dtype) {
     case INFINI_DTYPE_F16:
-        return _device_info->calculate<256, AddOp, half>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::AddOp, half>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_BF16:
-        return _device_info->calculate<256, AddOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::AddOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F32:
-        return _device_info->calculate<256, AddOp, float>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::AddOp, float>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F64:
-        return _device_info->calculate<256, AddOp, double>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::AddOp, double>(_info, workspace, output, inputs, stream);
     default:
         return INFINI_STATUS_BAD_TENSOR_DTYPE;
     }

src/infiniop/ops/add/cuda/add_cuda.cuh → src/infiniop/ops/add/nvidia/add_nvidia.cuh

@@ -3,6 +3,6 @@
 
 #include "../../../elementwise/cuda/elementwise_cuda_api.cuh"
 
-ELEMENTWISE_DESCRIPTOR(add, cuda, cuda)
+ELEMENTWISE_DESCRIPTOR(add, nvidia, cuda)
 
 #endif // __ADD_CUDA_API_H__

src/infiniop/ops/add/operator.cc

@@ -6,7 +6,7 @@
 #include "cpu/add_cpu.h"
 #endif
 #ifdef ENABLE_NVIDIA_API
-#include "cuda/add_cuda.cuh"
+#include "nvidia/add_nvidia.cuh"
 #endif
 
 __C infiniStatus_t infiniopCreateAddDescriptor(
@@ -31,7 +31,7 @@ __C infiniStatus_t infiniopCreateAddDescriptor(
         CREATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CREATE(INFINI_DEVICE_NVIDIA, cuda);
+        CREATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -46,14 +46,14 @@ __C infiniStatus_t infiniopGetAddWorkspaceSize(infiniopAddDescriptor_t desc, siz
 #define GET(CASE, NAMESPACE)                                                               \
     case CASE:                                                                             \
         *size = reinterpret_cast<op::add::NAMESPACE::Descriptor *>(desc)->workspaceSize(); \
-        return INFINI_STATUS_SUCCESS;
+        return INFINI_STATUS_SUCCESS
 
     switch (desc->device_type) {
 #ifdef ENABLE_CPU_API
-        GET(INFINI_DEVICE_CPU, cpu)
+        GET(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        GET(INFINI_DEVICE_NVIDIA, cuda)
+        GET(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
     default:
         return INFINI_STATUS_DEVICE_TYPE_NOT_SUPPORTED;
@@ -83,7 +83,7 @@ __C infiniStatus_t infiniopAdd(
         CALCULATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CALCULATE(INFINI_DEVICE_NVIDIA, cuda);
+        CALCULATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -99,7 +99,7 @@ infiniopDestroyAddDescriptor(infiniopAddDescriptor_t desc) {
 #define DELETE(CASE, NAMESPACE)                                                \
     case CASE:                                                                 \
         delete reinterpret_cast<const op::add::NAMESPACE::Descriptor *>(desc); \
-        return INFINI_STATUS_SUCCESS;
+        return INFINI_STATUS_SUCCESS
 
     switch (desc->device_type) {
 
@@ -107,7 +107,7 @@ infiniopDestroyAddDescriptor(infiniopAddDescriptor_t desc) {
         DELETE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        DELETE(INFINI_DEVICE_NVIDIA, cuda);
+        DELETE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:

src/infiniop/ops/clip/nvidia/clip_nvidia.cuh

@@ -2,7 +2,6 @@
 #define __CLIP_CUDA_API_H__
 
 #include "../../../elementwise/cuda/elementwise_cuda_api.cuh"
-#include "infiniop/ops/clip.h"
 
 ELEMENTWISE_DESCRIPTOR(clip, nvidia, cuda)
 

src/infiniop/ops/mul/cuda/mul_cuda.cu → src/infiniop/ops/mul/nvidia/mul_nvidia.cu

-#include "mul_cuda.cuh"
-#include "mul_cuda_internal.cuh"
+#include "mul_nvidia.cuh"
+#include "../cuda/kernel.cuh"
 
-namespace op::mul::cuda {
+namespace op::mul::nvidia {
 
 Descriptor::~Descriptor() = default;
 
@@ -43,17 +43,17 @@ infiniStatus_t Descriptor::calculate(
 
     switch (_dtype) {
     case INFINI_DTYPE_F16:
-        return _device_info->calculate<256, MulOp, half>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::MulOp, half>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F32:
-        return _device_info->calculate<256, MulOp, float>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::MulOp, float>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F64:
-        return _device_info->calculate<256, MulOp, double>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::MulOp, double>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_BF16:
-        return _device_info->calculate<256, MulOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::MulOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
     default:
         return INFINI_STATUS_BAD_TENSOR_DTYPE;
     }
 
     return INFINI_STATUS_SUCCESS;
 }
-} // namespace op::mul::cuda
+} // namespace op::mul::nvidia

src/infiniop/ops/mul/cuda/mul_cuda.cuh → src/infiniop/ops/mul/nvidia/mul_nvidia.cuh

@@ -3,6 +3,6 @@
 
 #include "../../../elementwise/cuda/elementwise_cuda_api.cuh"
 
-ELEMENTWISE_DESCRIPTOR(mul, cuda, cuda)
+ELEMENTWISE_DESCRIPTOR(mul, nvidia, cuda)
 
 #endif // __MUL_CUDA_API_H__

src/infiniop/ops/mul/operator.cc

@@ -7,7 +7,7 @@
 #endif
 
 #ifdef ENABLE_NVIDIA_API
-#include "cuda/mul_cuda.cuh"
+#include "nvidia/mul_nvidia.cuh"
 #endif
 
 __C infiniStatus_t infiniopCreateMulDescriptor(
@@ -32,7 +32,7 @@ __C infiniStatus_t infiniopCreateMulDescriptor(
         CREATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CREATE(INFINI_DEVICE_NVIDIA, cuda);
+        CREATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -47,14 +47,14 @@ __C infiniStatus_t infiniopGetMulWorkspaceSize(infiniopMulDescriptor_t desc, siz
 #define GET(CASE, NAMESPACE)                                                               \
     case CASE:                                                                             \
         *size = reinterpret_cast<op::mul::NAMESPACE::Descriptor *>(desc)->workspaceSize(); \
-        return INFINI_STATUS_SUCCESS;
+        return INFINI_STATUS_SUCCESS
 
     switch (desc->device_type) {
 #ifdef ENABLE_CPU_API
-        GET(INFINI_DEVICE_CPU, cpu)
+        GET(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        GET(INFINI_DEVICE_NVIDIA, cuda)
+        GET(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
     default:
         return INFINI_STATUS_DEVICE_TYPE_NOT_SUPPORTED;
@@ -84,7 +84,7 @@ __C infiniStatus_t infiniopMul(
         CALCULATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CALCULATE(INFINI_DEVICE_NVIDIA, cuda);
+        CALCULATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -108,7 +108,7 @@ infiniopDestroyMulDescriptor(infiniopMulDescriptor_t desc) {
         DELETE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        DELETE(INFINI_DEVICE_NVIDIA, cuda);
+        DELETE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:

src/infiniop/ops/sub/cuda/sub_cuda.cu → src/infiniop/ops/sub/nvidia/sub_nvidia.cu

-#include "sub_cuda.cuh"
-#include "sub_cuda_internal.cuh"
+#include "sub_nvidia.cuh"
+#include "../cuda/kernel.cuh"
 
-namespace op::sub::cuda {
+namespace op::sub::nvidia {
 
 Descriptor::~Descriptor() = default;
 
@@ -43,13 +43,13 @@ infiniStatus_t Descriptor::calculate(
 
     switch (_dtype) {
     case INFINI_DTYPE_F16:
-        return _device_info->calculate<256, SubOp, half>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::SubOp, half>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F32:
-        return _device_info->calculate<256, SubOp, float>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::SubOp, float>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_F64:
-        return _device_info->calculate<256, SubOp, double>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::SubOp, double>(_info, workspace, output, inputs, stream);
     case INFINI_DTYPE_BF16:
-        return _device_info->calculate<256, SubOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
+        return _device_info->calculate<256, cuda::SubOp, __nv_bfloat16>(_info, workspace, output, inputs, stream);
     default:
         return INFINI_STATUS_BAD_TENSOR_DTYPE;
     }

src/infiniop/ops/sub/cuda/sub_cuda.cuh → src/infiniop/ops/sub/nvidia/sub_nvidia.cuh

@@ -3,6 +3,6 @@
 
 #include "../../../elementwise/cuda/elementwise_cuda_api.cuh"
 
-ELEMENTWISE_DESCRIPTOR(sub, cuda, cuda)
+ELEMENTWISE_DESCRIPTOR(sub, nvidia, cuda)
 
 #endif // __SUB_CUDA_API_H__

src/infiniop/ops/sub/operator.cc

@@ -6,7 +6,7 @@
 #include "cpu/sub_cpu.h"
 #endif
 #ifdef ENABLE_NVIDIA_API
-#include "cuda/sub_cuda.cuh"
+#include "nvidia/sub_nvidia.cuh"
 #endif
 
 __C infiniStatus_t infiniopCreateSubDescriptor(
@@ -31,7 +31,7 @@ __C infiniStatus_t infiniopCreateSubDescriptor(
         CREATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CREATE(INFINI_DEVICE_NVIDIA, cuda);
+        CREATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -46,14 +46,14 @@ __C infiniStatus_t infiniopGetSubWorkspaceSize(infiniopSubDescriptor_t desc, siz
 #define GET(CASE, NAMESPACE)                                                               \
     case CASE:                                                                             \
         *size = reinterpret_cast<op::sub::NAMESPACE::Descriptor *>(desc)->workspaceSize(); \
-        return INFINI_STATUS_SUCCESS;
+        return INFINI_STATUS_SUCCESS
 
     switch (desc->device_type) {
 #ifdef ENABLE_CPU_API
-        GET(INFINI_DEVICE_CPU, cpu)
+        GET(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        GET(INFINI_DEVICE_NVIDIA, cuda)
+        GET(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
     default:
         return INFINI_STATUS_DEVICE_TYPE_NOT_SUPPORTED;
@@ -83,7 +83,7 @@ __C infiniStatus_t infiniopSub(
         CALCULATE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        CALCULATE(INFINI_DEVICE_NVIDIA, cuda);
+        CALCULATE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default:
@@ -107,7 +107,7 @@ infiniopDestroySubDescriptor(infiniopSubDescriptor_t desc) {
         DELETE(INFINI_DEVICE_CPU, cpu);
 #endif
 #ifdef ENABLE_NVIDIA_API
-        DELETE(INFINI_DEVICE_NVIDIA, cuda);
+        DELETE(INFINI_DEVICE_NVIDIA, nvidia);
 #endif
 
     default: