Merge branch 'dev' into AerialVision_cache_support

1 files changed, 12 insertions, 4 deletions

diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc
index ec68b5b..f7bb9cc 100644
--- a/src/cuda-sim/cuda-sim.cc
+++ b/src/cuda-sim/cuda-sim.cc
@@ -218,7 +218,7 @@ void gpgpu_t::gpgpu_ptx_sim_bindTextureToArray(const struct textureReference* te
    texInfo->Ty_numbits = intLOGB2(Ty);
    texInfo->texel_size = texel_size;
    texInfo->texel_size_numbits = intLOGB2(texel_size);
-   m_NameToTexureInfo[texname] = texInfo;
+   m_NameToTextureInfo[texname] = texInfo;
 }
 
 void gpgpu_t::gpgpu_ptx_sim_unbindTexture(const struct textureReference* texref)
@@ -226,7 +226,7 @@ void gpgpu_t::gpgpu_ptx_sim_unbindTexture(const struct textureReference* texref)
    //assumes bind-use-unbind-bind-use-unbind pattern
    std::string texname = gpgpu_ptx_sim_findNamefromTexture(texref);
    m_NameToCudaArray.erase(texname);
-   m_NameToTexureInfo.erase(texname);
+   m_NameToTextureInfo.erase(texname);
 }
 
 unsigned g_assemble_code_next_pc=0; 
@@ -1507,7 +1507,15 @@ static unsigned get_tex_datasize( const ptx_instruction *pI, ptx_thread_info *th
    std::string texname = src1.name();
 
    gpgpu_t *gpu = thread->get_gpu();
-   const struct textureInfo* texInfo = gpu->get_texinfo(texname);
+   /*
+     For programs with many streams, textures can be bound and unbound
+     asynchronously.  This means we need to use the kernel's "snapshot" of
+     the state of the texture mappings when it was launched (so that we
+     don't try to access the incorrect texture mapping if it's been updated,
+     or that we don't access a mapping that has been unbound).
+    */
+   kernel_info_t& k = thread->get_kernel();
+   const struct textureInfo* texInfo = k.get_texinfo(texname);
 
    unsigned data_size = texInfo->texel_size;
    return data_size; 
@@ -1918,7 +1926,7 @@ kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry,
                                              struct dim3 blockDim,
                                              gpgpu_t *gpu )
 {
-   kernel_info_t *result = new kernel_info_t(gridDim,blockDim,entry);
+   kernel_info_t *result = new kernel_info_t(gridDim,blockDim,entry,gpu->getNameArrayMapping(),gpu->getNameInfoMapping());
    unsigned argcount=args.size();
    unsigned argn=1;
    for( gpgpu_ptx_sim_arg_list_t::iterator a = args.begin(); a != args.end(); a++ ) {