Commit 46a15a1a authored by zhangyue's avatar zhangyue
Browse files

支持 elementwise float16 bfloat16 数据类型

parent 3a7633ba
...@@ -4,6 +4,7 @@ ...@@ -4,6 +4,7 @@
// This header file will only be include by .xpu file // This header file will only be include by .xpu file
#include "xpu/runtime.h" #include "xpu/runtime.h"
#include <xpu/kernel/xtdk.h> #include <xpu/kernel/xtdk.h>
#include <xpu/kernel/xtdk_bf16.h>
#include <xpu/kernel/xtdk_math.h> #include <xpu/kernel/xtdk_math.h>
#include <xpu/kernel/xtdk_simd.h> #include <xpu/kernel/xtdk_simd.h>
......
...@@ -10,6 +10,10 @@ private: ...@@ -10,6 +10,10 @@ private:
inline __device__ T sigmoid(T x) const { inline __device__ T sigmoid(T x) const {
return 1.0f / (1.0f + exp(-x)); return 1.0f / (1.0f + exp(-x));
} }
// float version of sigmoid
inline __device__ float sigmoidf(float x) const {
return 1.0f / (1.0f + exp(-x));
}
public: public:
// This static number must be set in other Ops // This static number must be set in other Ops
...@@ -21,9 +25,20 @@ public: ...@@ -21,9 +25,20 @@ public:
T out = gate * sigmoid(gate) * up; T out = gate * sigmoid(gate) * up;
return out; return out;
} }
// bfloat16 特化版本(使用 float 计算精度)
inline __device__ bfloat16_t operator()(const bfloat16_t *inputs) const {
float up_f = __bfloat162float(inputs[0]);
float gate_f = __bfloat162float(inputs[1]);
float out_f = gate_f * sigmoidf(gate_f) * up_f;
return __float2bfloat16(out_f);
}
} SwiGLUOp; } SwiGLUOp;
// __global__ template function instantiation
INSTANTIATE_ELEMENTWISE_KERNEL(SwiGLUOp::num_inputs, SwiGLUOp, float); INSTANTIATE_ELEMENTWISE_KERNEL(SwiGLUOp::num_inputs, SwiGLUOp, float);
INSTANTIATE_ELEMENTWISE_KERNEL(SwiGLUOp::num_inputs, SwiGLUOp, half);
INSTANTIATE_ELEMENTWISE_KERNEL(SwiGLUOp::num_inputs, SwiGLUOp, bfloat16_t);
} // namespace op::elementwise::kunlun } // namespace op::elementwise::kunlun
namespace op::swiglu::kunlun { namespace op::swiglu::kunlun {
...@@ -45,7 +60,7 @@ infiniStatus_t Descriptor::create( ...@@ -45,7 +60,7 @@ infiniStatus_t Descriptor::create(
const auto &up_shape = up_desc->shape(); const auto &up_shape = up_desc->shape();
const auto &gate_shape = gate_desc->shape(); const auto &gate_shape = gate_desc->shape();
CHECK_DTYPE(dtype, INFINI_DTYPE_F32); CHECK_DTYPE(dtype, INFINI_DTYPE_F32, INFINI_DTYPE_F16, INFINI_DTYPE_BF16);
CHECK_SAME_SHAPE(out_shape, up_shape, gate_shape); CHECK_SAME_SHAPE(out_shape, up_shape, gate_shape);
// create KUNLUN elementwise descriptor // create KUNLUN elementwise descriptor
...@@ -68,6 +83,10 @@ infiniStatus_t Descriptor::calculate( ...@@ -68,6 +83,10 @@ infiniStatus_t Descriptor::calculate(
switch (_dtype) { switch (_dtype) {
case INFINI_DTYPE_F32: case INFINI_DTYPE_F32:
return _device_info->calculate<8, op::elementwise::kunlun::SwiGLUOp, float>(_info, workspace, output, inputs, stream); return _device_info->calculate<8, op::elementwise::kunlun::SwiGLUOp, float>(_info, workspace, output, inputs, stream);
case INFINI_DTYPE_F16:
return _device_info->calculate<8, op::elementwise::kunlun::SwiGLUOp, half>(_info, workspace, output, inputs, stream);
case INFINI_DTYPE_BF16:
return _device_info->calculate<8, op::elementwise::kunlun::SwiGLUOp, bfloat16_t>(_info, workspace, output, inputs, stream);
default: default:
return INFINI_STATUS_BAD_TENSOR_DTYPE; return INFINI_STATUS_BAD_TENSOR_DTYPE;
} }
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment