fix: fix bug that tanh ptx require cuda11

NVIDIA · Aug 16, 2022 · d83204b · d83204b
1 parent e244b7b
commit d83204b
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 7 deletions.
diff --git a/docs/bert_guide.md b/docs/bert_guide.md
@@ -157,10 +157,7 @@ For those unable to use the NGC container, to set up the required environment or
 
     You can choose the tensorflow version and python version you want. Here, we list some possible images:
 
-    - `nvcr.io/nvidia/tensorflow:19.07-py2` contains the TensorFlow 1.14 and python 2.7. 
     - `nvcr.io/nvidia/tensorflow:20.12-tf1-py3` contains the TensorFlow 1.15 and python 3.8. 
-    - `nvcr.io/nvidia/pytorch:20.03-py3` contains the PyTorch 1.5.0 and python 3.6
-    - `nvcr.io/nvidia/pytorch:20.07-py3` contains the PyTorch 1.6.0 and python 3.6
     - `nvcr.io/nvidia/pytorch:20.12-py3` contains the PyTorch 1.8.0 and python 3.8
 
     To achieve best performance, we recommend to use the latest image. For example, running image `nvcr.io/nvidia/tensorflow:22.04-tf1-py3` by 

diff --git a/docs/decoder_guide.md b/docs/decoder_guide.md
@@ -156,10 +156,7 @@ For those unable to use the NGC container, to set up the required environment or
 
     You can choose the tensorflow version and python version you want. Here, we list some possible images:
 
-    - `nvcr.io/nvidia/tensorflow:19.07-py2` contains the TensorFlow 1.14 and python 2.7. 
     - `nvcr.io/nvidia/tensorflow:20.12-tf1-py3` contains the TensorFlow 1.15 and python 3.8. 
-    - `nvcr.io/nvidia/pytorch:20.03-py3` contains the PyTorch 1.5.0 and python 3.6
-    - `nvcr.io/nvidia/pytorch:20.07-py3` contains the PyTorch 1.6.0 and python 3.6
     - `nvcr.io/nvidia/pytorch:20.12-py3` contains the PyTorch 1.8.0 and python 3.8
 
     To achieve best performance, we recommend to use the latest image. For example, running image `nvcr.io/nvidia/tensorflow:20.12-tf1-py3` by 

diff --git a/src/fastertransformer/kernels/activation_kernels.cu b/src/fastertransformer/kernels/activation_kernels.cu
@@ -17,6 +17,11 @@
 #include "src/fastertransformer/kernels/activation_kernels.h"
 #include "src/fastertransformer/kernels/bfloat16_fallback_kenrels.cuh"
 #include "src/fastertransformer/utils/cuda_utils.h"
+
+#ifndef CUDART_VERSION
+#error CUDART_VERSION Undefined!
+#endif
+
 namespace fastertransformer {
 
 __forceinline__ __device__ float copysignf_pos(float a, float b)
@@ -28,7 +33,7 @@ __forceinline__ __device__ float copysignf_pos(float a, float b)
 
 __inline__ __device__ float tanh_opt(float x)
 {
-#if (__CUDA_ARCH__ >= 750)
+#if (__CUDA_ARCH__ >= 750 && CUDART_VERSION >= 11000)
     float r;
     asm("tanh.approx.f32 %0,%1; \n\t" : "=f"(r) : "f"(x));
     return r;