wehub-resource-sync
109 lines
4.7 KiB
C++
109 lines
4.7 KiB
C++
// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
#pragma once
|
|
#include <unordered_map>
|
|
#include "paddle/phi/kernels/xpu/xpu_api_wrapper.h"
|
|
|
|
namespace phi {
|
|
template <typename T>
|
|
inline XPUFCCalcType GetConvCalcType() {
|
|
std::unordered_map<const char*, XPUFCCalcType> calc_type_map = {
|
|
{"XPU_PADDLE_CONV_FLOAT16", XPUFCCalcType::FC_FLOAT16},
|
|
{"XPU_PADDLE_CONV_TF32", XPUFCCalcType::FC_TF32},
|
|
{"XPU_PADDLE_CONV_FLOAT", XPUFCCalcType::FC_FLOAT},
|
|
{"XPU_PADDLE_CONV_INT16", XPUFCCalcType::FC_INT16},
|
|
{"XPU_PADDLE_CONV_INT32", XPUFCCalcType::FC_INT32},
|
|
{"XPU_PADDLE_CONV_INT32_WITH_LL", XPUFCCalcType::FC_INT32_WITH_LL},
|
|
};
|
|
|
|
for (auto [env_str, calc_type] : calc_type_map) {
|
|
if (std::getenv(env_str)) {
|
|
return calc_type;
|
|
}
|
|
}
|
|
|
|
return FCCalcType<T>();
|
|
}
|
|
using XPUTypeFP16 = typename XPUTypeTrait<phi::float16>::Type;
|
|
using XPUTypeBF16 = typename XPUTypeTrait<phi::bfloat16>::Type;
|
|
template <typename QuantType>
|
|
struct XPUDefaultQuantType {
|
|
using Type = tfloat32;
|
|
};
|
|
|
|
template <>
|
|
struct XPUDefaultQuantType<XPUTypeFP16> {
|
|
using Type = XPUTypeFP16;
|
|
};
|
|
|
|
template <typename XPUType, typename QuantType>
|
|
struct XPUConvQuantTypeTrait {
|
|
using Type = QuantType;
|
|
};
|
|
|
|
#define DECLARE_XPU_UNSUPPORTED_QUANT_TYPE(TYPE, QUANT_TYPE) \
|
|
template <> \
|
|
struct XPUConvQuantTypeTrait<TYPE, QUANT_TYPE> { \
|
|
using Type = typename XPUDefaultQuantType<TYPE>::Type; \
|
|
};
|
|
|
|
#define XPU_CONV_FOR_EACH_UNSUPPORTED_QUANT_TYPE(_) \
|
|
_(float, XPUTypeFP16) \
|
|
_(XPUTypeFP16, int) \
|
|
_(XPUTypeFP16, int_with_ll_t) \
|
|
_(XPUTypeFP16, tfloat32) \
|
|
_(XPUTypeBF16, int16_t) \
|
|
_(XPUTypeBF16, int) \
|
|
_(XPUTypeBF16, int_with_ll_t) \
|
|
_(XPUTypeBF16, XPUTypeFP16) \
|
|
_(XPUTypeBF16, float)
|
|
|
|
XPU_CONV_FOR_EACH_UNSUPPORTED_QUANT_TYPE(DECLARE_XPU_UNSUPPORTED_QUANT_TYPE)
|
|
|
|
#ifndef PADDLE_WITH_XPU_XRE5
|
|
DECLARE_XPU_UNSUPPORTED_QUANT_TYPE(XPUTypeFP16, float)
|
|
DECLARE_XPU_UNSUPPORTED_QUANT_TYPE(XPUTypeFP16, XPUTypeFP16)
|
|
#endif
|
|
|
|
#define PD_PRIVATE_XPU_CONV_CASE(TYPE, calc_type, QUANT_TYPE, ...) \
|
|
case calc_type: { \
|
|
using TGEMM = typename XPUConvQuantTypeTrait<TYPE, QUANT_TYPE>::Type; \
|
|
__VA_ARGS__(); \
|
|
break; \
|
|
}
|
|
|
|
#define PD_VISIT_XPU_CONV_TYPES(TYPE, calc_type, func_name, ...) \
|
|
do { \
|
|
switch (calc_type) { \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_FLOAT, float, __VA_ARGS__) \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_TF32, tfloat32, __VA_ARGS__) \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_FLOAT16, XPUTypeFP16, __VA_ARGS__) \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_INT16, int16_t, __VA_ARGS__) \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_INT32, int, __VA_ARGS__) \
|
|
PD_PRIVATE_XPU_CONV_CASE( \
|
|
TYPE, XPUFCCalcType::FC_INT32_WITH_LL, int_with_ll_t, __VA_ARGS__) \
|
|
default: \
|
|
PADDLE_THROW(common::errors::InvalidArgument( \
|
|
"Function " #func_name " got invalid fc calc type %d", \
|
|
static_cast<int>(calc_type))); \
|
|
} \
|
|
} while (0)
|
|
} // namespace phi
|